The editor is clean. A script box on the left, a voice chip above it, playback at the bottom, and a control strip across the top for Smart Emotion, Emotion, Speed, Dynamics, Regenerate, Pronunciation and Pause. A new project asks for a name and a language mode (one language, or auto-detect), and then three product tour modals sit in a row before you reach the script. My default voice was Jackson. Clicking the voice chip opens a library of 739 characters, filtered by tags like #Characters, #Audiobook, #News, #Kid, #Anime, with a 'Preview with my script' button and a 'Clone Your Voice' banner at the top.

Some voices carry a 'Trial' badge. Jackson does. So do Alistair and Zara. Brandon, Alder, Felicity — the ones I looked at — do not. That badge turns out to matter a lot, but not until the download step.

I typed a 45-word script about a tool directory, clicked Regenerate, and three takes appeared in the generation history panel on the right about twelve seconds later. That part is genuinely free and genuinely fast: Typecast's pricing page says voice generation and playback are unlimited on every plan, and credits are only consumed by downloads. I played all three takes, applied Take 1, and edited the sentence in the script box to see if the timeline followed.

Then the download panel. Range is Full script only, because Selected sections is greyed out. Audio quality is Mid, because High is 44.1 kHz and carries a little crown, meaning Pro. Format is mp3, because wav is also greyed out. Output is Single file, because Split by sentence is greyed out too. And at the bottom of the panel it tells me exactly what my little script will cost: 239 credits.

Two hundred and thirty-nine characters in my script, two hundred and thirty-nine credits. One credit per character. My balance dropped from 3,000 to 2,761 when I confirmed it. The Free plan is 3,000 lifetime credits, and the pricing table calls that roughly five minutes of audio. I do not think that is right. My download was 14.076 seconds of speech for 239 credits, which puts the whole free pool at about 176 seconds — just under three minutes, not five.

The file itself is honest about what Free gives you. 16 kHz, mono, roughly 64 kbps, 112 KB for a 14-second line. That is phone-call quality, and if you are putting it under a video it will sound like it. The dialog also tells you that attribution is required on the Free plan and hands you the exact copy to use: Voice generated with Typecast (https://typecast.ai), with a one-click Copy button and a Download SRT file link next to it. I checked the mp3 for embedded metadata and there is none — no attribution tag inside the file — so carrying that credit is on you, not on the exporter.

I went looking for the parts of Typecast that make it different from a plain text-to-speech box, and the interesting walls are all about paid features leaking into a free workflow. Set Emotion to Happy and it applied — a toast said 'Paid feature applied. You can preview paid features free.' The take regenerated with the emotion on it. But at the same moment a red banner said the history had been reset because voice settings changed, and my three takes were gone from the panel. There was an Undo button. I needed it. Then I tried to download the emotional version and the panel said the audio included a paid feature, so download was available with the Pro plan: $29 a month.

I also swapped the voice to Brandon, one without the Trial badge, and the download asked for the Basic plan at $5 a month instead. Clicking Upgrade to Pro on any of these opens a Stripe checkout — Pro Plan, $29 per month, billed monthly — and I left that alone. For the record, Basic at $5 is the interesting price on this page, because it is where the 44.1 kHz downloads, the full voice library and the commercial licence live; Plus adds speed control at $19, Pro adds emotion and intonation at $29, and Business at $69 is for teams.

What I did not try: voice cloning (the Free plan has zero slots), the API, the video editor and talking avatar, and any non-English voice, including the auto-detect language mode. I also never listened to the emotion variants side by side, so I cannot tell you whether Happy actually sounds happy or whether it just sounds different in length. That is the one thing I would want before spending $29.

For anyone making short explainer or tutorial videos and willing to live with 16 kHz, a Trial voice and a visible Typecast credit, the free tier is enough to get a real file out. For anything you publish commercially, you are one click from a $5 plan anyway, so the sensible move is to stop trying to squeeze the free tier and just pick Basic.

Final rating: 3 out of 5.

Pixelcut's landing page pitches 'The AI Video and Image Platform' and plasters logos from Apple, Nike, and H&M across the header. Right below sits a direct link into their standalone Background Remover tool. No account required to start, no card requested, and a clean dropzone with three stock sample portraits waiting under the upload button. As someone who shoots and edits product photos regularly, I care less about the enterprise badge parade and a lot more about edge fringing on fabrics and whether I can get a file out of the tab without handing over a mailbox. I clicked the first portrait—a woman in an orange floral blouse against a cluttered room—just to time the server trip.

Four seconds to a 1080p checkerboard

The upload button spun for roughly 3.8 seconds before the whole page smoothly transitioned into an in-browser workspace at /ai-image-editor?tool=removeBackground. The background was gone. Not smeared, not haloed with that awkward white fringe you get on cheap segmentation models, but cleanly lifted right down to individual curls along her jawline and the shoulder seam of her shirt.

What caught me off guard was what happened when I clicked the black 'Download' button in the upper-right corner. Usually this is where the ambush happens. Instead, a clean two-row dropdown appeared. The bottom row said 'Full resolution 1280x1840 Free with sign in.' But the top row said 'Preview resolution 751x1080 Free.' I clicked it. The browser immediately saved a crisp, transparent PNG to disk with zero watermark, zero login prompt, and zero email capture.

751×1080 is not a throwaway thumbnail. That is taller than high-definition video and more than sharp enough for an Instagram story, an ecommerce catalog card, or a newsletter header. When you compare that to the 500-pixel mush most competitors hand over without an account, Pixelcut is actually giving away a genuinely usable production cutout. You can drop a quick product shot on it, grab the PNG, and paste it into Figma or Shopify without ever creating a login.

The canvas UI also gives you a side-by-side 'Compare original and result' toggle, zoom controls from 10% to 500%, and an option to cast a subtle contact shadow underneath the cutout. The shadow control worked instantly without lagging the viewport, grounding the subject instead of leaving her floating like a flat sticker.

Typing a prompt turns into a paywall

Once the cutout landed on the checkerboard canvas, the editor opened up a left-hand toolbar with secondary tools: Background, Retouch, Expand, Upscale, and Enhance. At the bottom of the canvas sat a prompt bar labeled 'Model AI Backgrounds.' If you sell physical goods online, this is the real feature you want: drop a raw cutout onto a clean marble kitchen counter or a warm studio table without setting up three softboxes in your living room.

I typed 'minimalist white wooden table with soft sunlight studio' into the text box and hit the black generate arrow. That was the end of the anonymous free ride. The canvas instantly blurred behind a full-screen modal titled 'Sign Up or Log In' with an internal illustration labeled 'Paywall.' The prompt didn't execute in preview mode; it didn't give me a single free watermarked test render. The moment you move past static segmentation into generative diffusion, the door shuts firmly.

The same gate applies to the other generative tools in that rail: generative expand, AI upscaling past standard resolution, and the virtual actor staging all slam into the exact same login prompt. Pixelcut draws a sharp line between its legacy utility features and its generative AI suite: mask cutting is a generous free funnel, while prompt synthesis is strictly gated behind registered accounts.

The Cloudflare barrier on email signup

Generative models cost real GPU compute, and nobody hands out unlimited image synthesis to anonymous web crawlers. So I went to create an account. The dialog offered Continue with Google, Continue with Apple, and Continue with email. I clicked email to keep things isolated from my main personal accounts.

I typed in forleoliang@hotmail.com. Right below the text field, an iframe containing a Cloudflare security challenge popped into the frame. The blue 'Continue' button switched to disabled, and there it stayed. The Turnstile widget hung, waiting on verification inside the modal, leaving the submit button dead in the water. If you don't want to use Google or Apple OAuth, that email flow can be a sudden dead end.

I didn't test their bulk editor on hundreds of catalog SKUs, nor did I burn through the 600 monthly credits on their Pro tier ($8 to $10 a month depending on annual billing) to see how their AI actor avatars handle hands or complex jewelry reflections. But as a fast, browser-based utility, Pixelcut solves the single biggest annoyance with online background removers: you can walk up to the URL, drop a portrait or product shot on the canvas, and walk away with a clean 1080p cutout in four seconds without handing over your inbox.

Final rating: 4 out of 5.

I tested them on different days with different questions, so this is not a same-prompt shootout. What I can compare is what each free path let me finish, and where each one stopped short of a five.

A three-part query with a lie planted in it

For Felo I stayed logged out on purpose. The query bundled a capital, a UNESCO listing, and a deliberate false premise: the capital of Argentina, whether Iguazu is a World Heritage Site, and whether Mount Aconcagua is in Chile.

About 21 seconds later the lead paragraph got all three right. Buenos Aires as capital. Iguazu inscribed in 1984 on the Argentine side and 1986 on the Brazilian side. Aconcagua came back with a bold not — Mendoza province, roughly 15 km from the Chilean border, not Chile. The panel showed 30 Source & Thinking. Wikipedia Buenos Aires and Britannica Buenos Aires both opened cleanly. The UNESCO World Heritage Centre URL was genuine; Cloudflare threw an interstitial before the page appeared.

The refusal is the part I would keep. Plenty of tools will happily answer "where in Chile is Aconcagua" as if the question were sound.

The same answer rewritten in Spanish, still anonymous

Still anonymous, I asked Felo to rewrite the result in Spanish and keep the citations. About 31 seconds later I had a full Spanish answer with the reference markers intact, and the Chile trap was still refused in the new language. The source count climbed from 30 to 41.

That is the multilingual case in one move: not just translating a summary, but holding the citation structure and the fact-check across languages. The longer walkthrough is in my Felo review.

Where Felo stops: a captcha before creation

Anonymous search held up. Creation sat behind a login, so I tried the email signup path. The verification code arrived in about 55 seconds — in Junk, with a 60-second on-screen countdown. After I entered the code, the page threw a reCAPTCHA image grid: select all squares with crosswalks. I do not solve those in these sessions. Skip returned "Verification failed." No account. Clicking Generate AI Slide produced a login wall: Google, Apple, or email.

So the honest read is that the anonymous path is the product I actually used, and it worked. The moment I wanted slides, an image captcha stood in the way.

Pricing was readable while logged out. Free forever includes 200 bonus credits a day, unlimited high-speed search, 5 professional searches a day, and 3 file analyses a day. Pro runs $14.99 a month, or $12.50 a month billed annually at $149.99 a year.

Seven seconds on fasting and HbA1c

My Perplexity session ran separately, earlier, in an Incognito Free window. The UI said sessions are not saved and expire after 24 hours. No signup and no card for the queries I ran.

The first question was dense: what 2023–2024 systematic reviews and randomized trials found about intermittent fasting and HbA1c in adults with type 2 diabetes, with studies cited and limitations called out. First useful prose landed in about seven seconds with 24 sources, and it carried real numbers — a 2023 RCT meta-analysis SMD of −0.08 (95% CI −0.20 to 0.04; P=.19), INTERFAST-2 at −7.3 mmol/mol (~−0.6 percentage points) versus control +0.1, EARLY 5:2 meal-replacement −1.9 points at 16 weeks versus metformin −1.6 and empagliflozin −1.5. A soft "This question looks difficult" card with Upgrade to Pro sat over the answer. The answer still finished on Free. The full session is in my Perplexity review.

A table rebuilt in eight seconds

I asked it to turn the evidence into a comparison table across the 2023 meta-analysis, INTERFAST-2, EARLY, and a 2024 16:8/14:10 trial, and to flag bundled interventions. About eight seconds later, 30 sources deep, the table held the caveats: insulin-dose adjustment in INTERFAST-2; meal replacements, calorie guidance, contact, and education bundled into EARLY. That was the moment I stopped testing "can it write" and started trusting it for structure.

Thirty-six seconds, then Chinese

A second query — remote work and software-engineer productivity from 2023–2025 studies, with causal versus correlational evidence — took about 36 seconds and produced serious synthesis across 25 sources: Gibbs/Mengel/Siemroth (2023) full-remote IT setting with output/hour down about 8–19%; Bloom/Han/Liang (2024 Nature) hybrid RCT, two days a week at home, no significant performance damage and lower attrition; BLS industry evidence labeled correlational. The Nature citation opened the real 12 June 2024 paper.

The wobble was language. The prompt was English; the whole answer rendered in Chinese. The retrieval path still worked. The language router did not. Another Upgrade to Pro card and a "Pro Free preview of advanced search enabled" banner sat alongside it.

What the pricing pages said without an account

On annual Personal pricing, Perplexity listed Pro at $17 a month and Max at $167 a month. Free was referenced as the base tier ("Everything in Free"), but the modal showed no separate numeric Free allowance. Felo's logged-out page was more forthcoming with the free daily caps and the two Pro prices above.

Pick by the job, not the brand

If I need a quick geographic or factual trap check — something with a planted false premise — and I want it rewritten in another language without losing citations, I open Felo and stay anonymous. That path worked, including Spanish on the first try. Budget patience if you need slides: signup hit an image captcha for me.

If I need literature synthesis with numbers, caveats, and a table I can rebuild, I open Perplexity. A Free Incognito session delivered all of that without an account. Watch for soft Pro upsells, an unexpected language flip on a hard prompt, and a pricing modal that will not show you a Free meter.

Both earned a four. Pick the shape of answer you actually need, then open the citations before you trust the sentence they sit under.

The file was a made-up 40-row CSV of camera-accessory orders (order ID, date, product, country, quantity, unit price, revenue) across July to September 2026. I kept the correct answers for that file in a separate document before I opened the Powerdrill listing, so every row below is graded against those, not against my impression of the replies.

Getting in

"Start for free" asked for an email and nothing else. No password, no Google button needed, no card. A six-digit code arrived in about 20 seconds, in the Junk folder of my Hotmail address, from hello@powerdrill.ai. After that there was no onboarding quiz at all. It dropped me straight into a "Default Workspace" with a crown icon and 1,000 next to it, which is the daily credit counter. A "Super Sale Week" banner sat across the top. I ignored it and uploaded the file through the paperclip menu. It showed up as a 1.68 KB chip above the prompt box.

One thing to know before you sign up: the product is now called Powerdrill Bloom, and it is not a spreadsheet. There's a "Spreadsheets & Data" shortcut on the start screen, so I half expected a grid. What you actually get is a chat window with a hidden Linux sandbox behind it. The agent writes awk or Python, runs it against your file, and talks back.

The scorecard

Nine questions in a row, all on the default Lite model. The first seven are a clean sweep. Not "close enough", exact. The eighth is right too. The last row is the asterisk.

Short takes, row by row

Total revenue, $3,318. Right, and it threw in the per-order range ($9 to $236) without being asked. 41 seconds, which felt slow for a sum, until I opened the "Analysis process" drawer and saw why. It first tried to pipe the column into bc, got bc: command not found (exit code 127), and rewrote the line in awk without any fuss. I liked seeing that more than I expected. The answer wasn't a guess from a language model. It came from a script I could read.

By month. Right on all three, July $1,372, August $935, September $1,011. It also drew an interactive bar chart with hover values and saved it as a downloadable HTML file. Slowest answer of the session at 59 seconds.

Top product by revenue / by units. Black Mist 1/4 at $1,180, Lens cloth at 21 units. Both right, and it pointed out the cheap lens cloth tops volume but comes last in revenue. A fair observation, not a profound one.

Average order value, $82.95. Right. 3,318 over 40.

US revenue in September, $71. This was my trap question, two filters at once, and the kind of thing chat tools often fumble by summing the whole month. The drawer showed a single awk line matching 2026-09 and US and printing 71. It added that the US was about 7% of September. True.

Revenue by country chart. Japan, Germany, Argentina, United States, all four bars matched my numbers. It expanded my country codes into full names on its own.

The part where I tried to change a number

I told it order O1001 was wrong and should be $1,000 higher. The first reply did the arithmetic in text ($4,318 total, $2,372 for July), which was correct. It then asked whether I wanted the CSV itself updated. I said yes. It wrote a short Python script that opened the uploaded file, set O1001's revenue to 1236, and wrote it back. It printed "CSV file updated" and a new total of $4,318.

Fine so far. Then I asked for a snippet that totals revenue where quantity is 3 or more, and to run it. It wrote clean code, ran it, and reported $2,172 across 20 orders, "about 65% of your total $3,318 revenue."

O1001 has a quantity of 4. If that edit had held, the answer should have been $3,172, out of a $4,318 total. $2,172 is exactly what the untouched file gives. So between one turn and the next, my correction was gone. Either the edit never persisted into the next run, or it read a fresh copy of the upload. I can't tell which from the outside, and Powerdrill didn't notice anything either. It simply reported the old total as if nothing had happened two messages earlier.

That's the problem with a chat-only data tool. In a spreadsheet you'd see the cell. Here, you only find out if you already know the answer.

What it costs

Nine questions, several charts, a file rewrite, and the counter still read 1,000 at the end. Lite didn't touch the credits at all, at least for me. The free plan advertises 1,000 daily refreshed credits, Excel/CSV/PDF uploads, charts and summaries, and one scheduled task. Pro is $16.58 a month, or $13.27 billed yearly ($159.20 the first year, then $199). Plus is $33.25 a month, and Premium jumps to $165.83. There's also a one-off $5 pack for 1,000 extra credits.

Who should bother

If you have a CSV and a few questions, and you're the kind of person who'll open the drawer and glance at the awk line, Powerdrill is quick, free, and honest about its working. I'd point a shop owner or a marketer at it before I'd point them at a pivot table tutorial. If you need to edit data and keep asking questions of the edited version, I wouldn't trust it without re-uploading the corrected file yourself. And if you wanted a real grid with formulas, this isn't that product, whatever the start-screen shortcut suggests.

I didn't try Excel or PDF files, the paid Claude-Code-level agent tier, scheduled tasks, or the open-data search. All of my testing was on one small CSV in one evening.

Final rating: 4 out of 5.

I tested them on different days with different questions, so this is not a same-prompt shootout. What I can compare is what each free path let me finish, and where each one stopped short of a five.

Paris in 1839, with the papers open

Signing up for Paperguide took under about two minutes. Email sign-up at paperguide.ai, a Turnstile check that passed on its own, no verification mail, and a pricing modal I dismissed without a card. Then I was on the AI Search dashboard.

My query was about the daguerreotype: when and where the process was publicly announced, and what evidence historians cite for its reception in 1839. The answer landed on Paris in the summer of 1839, with numbered inline cites. I opened at least three of them. Adamson 1989 in *History of Photography* (DOI 10.1080/03087298.1989.10441188), a 1939 *Nature* note (DOI 10.1038/143963a0), and Stănescu 2022 (DOI 10.24193/subbhist.2021.1.08) all matched title and DOI. The Stănescu abstract itself says the invention was presented in Paris in the summer of 1839 by Louis Daguerre. That is the moment the product has to earn: not a tidy summary, but a cite you can click and confirm.

I tried a trap next: did Louis Daguerre patent the daguerreotype in London in 1835, and was Talbot involved? Paperguide refused correctly. It pointed to the French state acquisition in 1839, Arago's presentation on 19 August 1839, English rights going to Richard Beard, and Talbot's separate calotype work. No fabricated 1835 London patent cites. For a short photo-history note, that refusal mattered as much as the first answer. The longer walkthrough is in my Paperguide review.

One Adamson paper that stayed put

I saved the Adamson paper into My References. After a navigate to /references/, All Papers still showed 1. That is a small thing, and it is the kind of small thing that decides whether I trust a free research tool with a note I care about.

The free meters are less tidy. After two AI Searches the counters sat at ai_search 2/20 and dailyAiGenerations 2/5, while the plan credits stayed 1500/1500 with usedCredits at 0. AI Search burns a separate search quota, not the marketed 1,500 AI credits. The free plan ($0) lists those 1,500 credits for Reports and Writer work, plus limited Reference Manager and limited AI Searches. Plus is $12 per seat per month billed annually; Pro is $29; Evidence Teams is $89. If you came for the big credit number, check which meter your search actually hits before you plan a week's worth of queries around it.

A meter across twenty papers

Consensus is built for a different shape of question. Mine was: does intermittent fasting improve insulin sensitivity in adults? Logged-out submit opened an email-auth wall (email, Google, or Apple). I used the email-code sign-up path — no card asked for. After onboarding, the sidebar showed 5/5 Pro messages left.

About thirteen seconds later the Pro synthesis was populated. The results pool sat at "52.3K". A "Read Abstracts and PDFs 20" line framed the working set. The AI synthesis opened with a clear yes — intermittent fasting improves insulin sensitivity in adults, though effects vary — and the Consensus Meter (N=20) made the vote visible: roughly 15 Yes, 3 Possibly, and 2 other. Paper cards carried titles, years, journals, and PDFs. Study Snapshot was free, with "10 Study Snapshots left" and a refresh date. The first paper I opened had DOI 10.1155/2022/6999907.

What kept it from a five is the same wall that starts the session. You cannot see results before the email step. The filter left "Results 52.3K" unchanged in my run. Snapshot stopped at 4 of 7 attributes. And Pro messages are a hard allowance once you are inside. The full session notes are in my Consensus review.

Two free paths that run out differently

On annual pricing as shown in the app, Consensus Free is $0 with basic paper search free and limited Pro messages — my session started at 5/5 and dropped to 4/5 after one search. Pro is $12/month. Deep is $45/month. Paperguide Free is $0 with the credit and search split described above; its paid seats start at $12 annually for Plus.

Neither tool replaces a citation-graph product like Connected Papers when you already have a seed paper and want neighbors. That job lives elsewhere, and Consensus already has that comparison on the site. Here the choice is simpler. Need a yes/no science question with a meter across papers after you can sign in? Open Consensus. Need a grounded narrative answer with openable citations, and a free path that reaches search without verifying email first? Open Paperguide. Both earned a four. Pick the shape of answer you actually need, then open every DOI before you trust the sentence it sits under.

The signup was email and a password, no Google, no card. Cloudflare's Turnstile sat on the form and auto-cleared without a picture puzzle. No verify email. A pricing modal and a RESEARCH50 coupon banner showed up immediately; I closed both. Under two minutes later I was staring at "Hello Researcher, What would you like to research today?" with AI Search selected.

The first suspect: ungrounded chat

Most AI research tools will talk about 1839 whether or not they have a paper. Paperguide's pitch is the opposite — every claim linked to a source. So I typed: "When and where was the daguerreotype process publicly announced, and what evidence do historians cite for its reception in 1839?"

It thought for a bit, then wrote that the process was publicly announced in Paris in the summer of 1839 by Louis Daguerre, and that historians mostly track reception through contemporary newspaper reports rather than one official text. Inline blue numbered cites sat on the sentences. A honesty note in the answer itself admitted that none of the papers it retrieved pinned the French Academy / Arago date. That note is the early clue. Keep it.

Opening the papers

I clicked through three of them.

Adamson, "1839 — The year of Daguerre," History of Photography, 1989, DOI 10.1080/03087298.1989.10441188. The in-app drawer matched the title. The Taylor & Francis page opened. It's a newspaper-report survey of that year — exactly the evidence strand the answer leaned on.

Nature's "A Hundred Years of Photography" at DOI 10.1038/143963a0. Real centenary piece. Plausible for process details.

Stănescu, "From Daguerreotype to Autochrome…," Studia Universitatis Babeș-Bolyai Historia, 2022, DOI 10.24193/subbhist.2021.1.08. Open access. The abstract literally says the daguerreotype was brought to public attention in the summer of 1839 in Paris by Louis Daguerre. Title match, claim match.

So the "will invent DOIs" suspect is gone for this session. The papers exist, the titles match, and at least two of them say what the answer used them for.

Planting a false patent

I planted a lie on purpose: "Did Louis Daguerre patent the daguerreotype in London in 1835, and was Talbot involved in that patent?"

Paperguide said no. French state acquisition in 1839, Arago's lecture on 19 August 1839, English rights going to Richard Beard, Talbot working a separate calotype patent. No 1835 London patent appeared in its sources, and it said so. That is the payoff of the earlier honesty note — the first answer left the Academy date soft, then the follow-up produced the date with the corrective history instead of inventing a patent to please me.

A saved paper, and a counter that won't move

I hit My References on the Adamson paper, then navigated to /references/. It was still there under All Papers (1), Added Oct 25, 2026, PDF icon attached. So the "Reference Manager is vapor" suspect is out.

The credit counter is the odd one. Before the searches I had 1,500 of 1,500 AI credits and 0 of 20 AI Searches used. After two AI Searches: still 1,500 credits remaining, zero used — but AI Searches sat at 2 of 20, and daily AI generations at 2 of 5. Free marketing sells "1,500 AI credits per month" for Agents, Reports, and Writer. Search burns a separate limited quota. If you came for Search, the big number on the Free card is not the meter that moves.

What free actually costs

Pricing (yearly toggle, save 41%, RESEARCH50 flash sale running while I was there): Free is current plan — 1,500 AI credits/mo, limited Reference Manager, limited AI Searches, AI Writer, Chat with PDF, Literature Review and Deep Research Report, Zotero/Mendeley import. Plus is $12/seat/mo billed annually (12,500 credits, unlimited searches and Reference Manager). Pro $29. Evidence Teams $89. I didn't buy anything.

I didn't run Systematic Review, Extract Data, AI Writer, Chat with PDF on an upload, or a Zotero sync. Free Reference Manager also left some export/cite buttons grayed on certain titles.

If you need a cited answer on a history question and will open the papers yourself, Paperguide's free AI Search did the job — and it would not invent an 1835 London patent when I asked it to. If you need twenty-plus searches a month, a full bibliography export, or a credit meter that matches the Free card, the free plan is a short demo. More details and alternatives on Paperguide on ToolifyNav.

Final rating: 4 out of 5.

That moment is most of the answer. Both Jenni and Paperpal get sold as "AI academic writing". In my tests they did opposite jobs. Jenni wrote text I then had to check. Paperpal checked text I had already written. If you have a blank page and you're willing to open every source, start with Jenni. If the draft exists and needs to read like journal English, use Paperpal. Writing a paper this week, I'd use both in that order, and I'd verify the sources between the two steps. I haven't run that whole pipeline end to end; it's simply what the two tests point to.

One caveat before the details. I tested them on different days with different text, so this isn't a same-paragraph shootout. What I can compare is what each one did with the job it was built for, and what each free plan let me finish.

Four papers, three that held up

Signing up for Jenni (4/5) took about two minutes. Email sign-up, no card, no phone, no captcha. The verification mail went straight to Junk. There was one onboarding question about my role, and then I was in the editor.

My prompt was "How the daguerreotype changed photography". I accepted five autocomplete suggestions. Most came in around 4 seconds and one took about 10. Together they made a 199-word paragraph with five in-text citations drawn from four sources: Balbás et al. 2025, Männig 2021, Keim 1965 (cited twice) and Bright 1992.

I opened all four. Every one was a real paper with the title Jenni gave it. The Balbás abstract calls daguerreotypes the first widely disseminated photographic process, which supports the 1839 sentence. Keim's "Photography and Reality" backs the line about being able to count the tiles on a roof. Männig held up, as above. Bright's "The Machine in the Garden Revisited" exists and the title is right, but I couldn't confirm it supports the sentence it's attached to. So three of four clearly back their claims, and the fourth is unconfirmed rather than invented.

What it left out mattered more to me than what it put in. There was no Paris, no Niépce dying in 1833 and no Talbot patenting the calotype in 1841. The dates it did use were right. When I steered it with "Nicéphore Niépce, Daguerre's earlier partner", it wrote about 100 accurate words in roughly 5 seconds: pewter heliographs in the 1820s, the latent image on a silver-coated copper plate, the one-of-a-kind positive. The source count went from four to eight.

Paraphrase took about 4 seconds and cut the first paragraph from 199 to 163 words with the citations still attached. Once, after the tab had sat idle, a "Reconnecting… Editing is paused" notice appeared, which is the kind of thing you'd want to know before a deadline.

The bibliography you can't read

Here's the catch on Jenni's free plan. The reference list at the bottom of the document is blurred, with this note: "References are a paid feature. Upgrade to view, copy, and export references." Switching from APA to MLA was instant and the heading changed to "Works Cited", but the entries stayed blurred. The free .docx export downloaded fine at 17 KB. It has Word citation fields in the body and no reference list on the page.

The free limits are tight, too: 10 autocompletes a day, 3 AI edits, 5 chat messages and 10 PDF uploads. A paragraph and a half used 8 of the autocompletes and 2 of the edits. Plus is $12/month billed annually or $30 month to month, and that's the tier that unlocks the references. The full test is in my Jenni review.

Seven slips, eleven cards

Paperpal (4/5) never wrote a sentence for me, because I didn't ask it to. Signup went through an Editage sign-in page (same company, shared accounts). There was no verification email, no card and a three-question quiz. The public pricing URL gave me a CloudFront 403 error page, but the plans page inside the app loaded fine.

I gave it an 86-word paragraph I'd seeded with seven grammar slips: subject-verb agreement, "you're" for "your", "their is" and similar. It came back with 11 suggestions: 6 grammar, 3 readability, 1 mechanics and style, 1 vocabulary. It caught all seven, and none of the suggestions was locked for paying users. Two of those basic errors were filed under a vague "Rephrase this portion for readability" card, which is a strange label for a plain grammar mistake. One of the three extra suggestions changed "keep" to "maintain". That's the academic pull at work: correct, and a little stiffer than what I'd written. A second, 35-word paragraph with three slips went 3 for 3.

"Make Academic" on one broken sentence took about 2.5 seconds, fixed both errors and kept the meaning. The daily counter went from 5/5 to 4/5 with no upgrade wall. The free document check took about 80 seconds on the short paragraph: 0% AI detection and 0% plagiarism similarity. Reference Check sits behind Prime. More detail is in my Paperpal review.

The meters count different things

The two free plans run out in different ways. Jenni counts what you ask for: autocompletes per day and a small fixed pool of AI edits. Paperpal's 200 monthly grammar edits are charged when suggestions are generated, not when you accept them. The counter dropped from 200 to 189 as soon as the 11 cards appeared. Accepting them cost nothing, and the second paragraph took it to 186. If you paste a messy 3,000-word chapter, the month could be gone before you've decided which fixes you agree with.

On price they start in the same place:

- Jenni Plus: $12/month billed annually, $30 month to month. Pro is $29 annually or $73 monthly. - Paperpal Prime: $12/month billed annually, $25 month to month. Pro is $29 annually or $59 monthly.

Both have a free rewrite button. Jenni gives you 3 AI edits in total; Paperpal gives you 5 AI actions a day. Neither of my tests covered the other tool's main job. I didn't grade Jenni on catching my grammar, and I didn't ask Paperpal to generate citations. So read this as a choice between jobs, not a ranking. The side-by-side page lays out the listing details.

Which one to open first

If the page is empty, open Jenni, write the claim you need, then click every source before you believe it. Three of my four checked out. That's good, and it still means one sentence had support I couldn't confirm. If you're on the free plan, budget for copying the references by hand or for the $12 annual tier.

If the words are already yours and the problem is how they read, open Paperpal. Run your cleanest version through it, because every suggestion it generates counts against the meter. Then take its "maintain"-style changes on a case-by-case basis.

Getting there was quick. I signed up with my Hotmail address and a password, no card, no phone, no picture puzzle. The verify email showed up in Junk within seconds, which is a little embarrassing for a company selling to universities, but it worked. One onboarding question ("What is your current role?"), I skipped the tour, and I was in the editor about two minutes after clicking "Start writing – it's free."

Five suggestions and a paragraph

I wasn't writing anything. I just kept accepting. Five suggestions, mostly around four seconds each, one closer to ten, and the counter in the corner said 199 words. One paragraph, five in-text citations from four sources: Balbás et al. 2025, Männig 2021, Keim 1965 twice, and Bright 1992. A sixth suggestion started coming and I hit Escape.

It reads like a decent first-year grad student who has read the abstracts. "Count the tiles on a roof" is in there as a quote. The dates it used were right, publicly introduced in 1839. It also never mentioned Paris, never mentioned that Niépce died in 1833, never got to Talbot's calotype patent in 1841. I had those three facts ready to catch it lying. It didn't lie, it just didn't go there.

Opening every source

This was the part I actually cared about, because cited AI writing usually falls apart here. The "Sources" button next to the citation style dropdown opens a panel with all four papers, impact factor tags, and a View or Open Full Text link on each.

All four were real. Balbás et al. opened on the Firenze University Press journal site, title matching, and its abstract literally calls daguerreotypes "the first widely disseminated photographic process made public." That's the sentence Jenni used it for. Keim's "Photography and Reality" opened on Cambridge Core, and the roof-tiles line is right there in the abstract. Good.

Männig's paper is about tableau vivant and social media, so I expected a stretch. Open Full Text loaded the whole 24-page open-access PDF inside Jenni. On pages 135–136 it quotes Poe in 1840 calling the daguerreotyped plate more accurate than "a positively perfect mirror." That's the mirror claim Jenni hung on it. I was wrong, it fit.

Bright's "The Machine in the Garden Revisited" opened on Taylor & Francis, title correct. Whether it supports a sentence about the daguerreotype's scientific credibility, I can't tell you. There's no abstract on the record and it reads like a paper about landscape photography. So: four real papers, three that clearly back their sentence, one I'd want to read before submitting anything.

Pushing it toward Niépce

To see if it would make up history once I steered, I started a new paragraph with "Nicéphore Niépce, Daguerre's earlier partner," and waited. About five seconds. It added around 100 words: heliographs on pewter in the 1820s, long exposures, a latent image on a silver-coated copper plate, a one-of-a-kind positive you couldn't copy like later paper negatives. All fine. Sources jumped from 4 to 8. Neuerburg and Slijkhuis 2013 and Yetisen et al. 2015 are real papers. I didn't chase the other two (Richter and Gernsheim, Fuente).

Still no 1833, still no Talbot. If your essay needs those specifics, you type them yourself.

Paraphrase, and a sync wobble

I selected just the first Niépce sentence and opened the AI edit menu: Improve fluency, Paraphrase, Simplify, Strengthen argument, Change tense, a few more. Paraphrase took about four seconds and gave me "Before Daguerre, his former partner Nicéphore Niépce made heliographs on pewter plates in the 1820s. Long exposure times and fragile results made them impractical until Daguerre refined the process for commercial use." Cleaner than the original. Counter went 266 to 260.

Right as the menu opened, a box popped up at the top: "Reconnecting… Editing is paused while we reconnect." I'd left the tab sitting for a few minutes before that. It came back, but if that happens mid-thought in a long session, it's annoying.

Earlier I'd paraphrased the entire first paragraph the same way. 199 words became 163 and all five citations stayed attached to the right sentences.

The blurred box at the bottom

Under the text there's a References section. On free, it's blurred, with a banner: "References are a paid feature. Upgrade to view, copy, and export references." You can see the shapes of the entries but not copy them.

Switching APA 7th to MLA 9th worked instantly. In-text cites dropped the years, the heading changed to "Works Cited," and it was still blurred.

Then I exported to Word, which is free. A 17 KB .docx with my paragraphs and native Word citation fields, but no reference list on the page. The source details are tucked inside the file's citation data, so Word could probably build a bibliography from them. Jenni won't do it for you unless you pay.

What free actually gets you

There's no daily word cap. The number in the corner is just document length. The real limits are in the account menu: 10 AI autocompletes a day, 3 AI edits, 5 AI chat messages, 3 reviews, 3 workflow runs, 10 PDF uploads. By the end I'd used 8 of 10 autocompletes and 2 of 3 edits. That's one paragraph and a half. Free is a demo, not a writing plan.

Plus is $12 a month billed annually, $30 if you pay monthly, for 5,000 autocompletes, 500 edits and full document export with references. Pro is $29 annually or $73 monthly, unlimited. That monthly jump is steep.

I didn't upload my own PDFs, didn't touch the chat panel, Review, workflows, LaTeX export or the Chrome extension.

For a student writing a literature-heavy background section who will actually open every source, Jenni is useful, because the citations it gave me were real and mostly honest. If you want a finished bibliography on free, or you need it to know which facts matter, look elsewhere. More details and alternatives on Jenni on ToolifyNav.

Final rating: 4 out of 5.

Official sites for the four tools: chatpdf.com, humata.ai, explainpaper.com, scisummary.com.

That moment is roughly why I wanted to put four tools side by side. I've now tested four apps that promise to get you out of a PDF faster, and they turned out to be doing four different jobs. ChatPDF and Humata let you ask a whole file questions and click back to the source. Explainpaper explains whatever sentence you highlight. SciSummary writes a summary and keeps its chat behind a trial. I tested them on different days with different documents, so this isn't a same-file shootout. It's what each one did when I gave it something I already knew the answers to.

The steward hours that lost their page

For ChatPDF I wrote a two-page handbook for a darkroom that doesn't exist, "Riverside Film Lab," and spread the facts across both pages on purpose. Upload worked without an account. About 15 seconds later it said "All 2 pages checked, ready to chat!" and greeted me with "Hey there, film enthusiast!", which I could have done without. Signing up later took a plain email form, with no captcha, no verification mail and no card, and the chat came along.

Answers took two or three seconds each. The page-two developer rule (1+9, thrown out after 20 prints per litre or 8 hours) came back correct, and the ↗2 chip jumped to page two and lit the sentence in blue. Then the cross-page one above: right answer, and it also caught that the lab is shut on the day I asked about, but the steward rule never got its chip.

The trap mattered more to me. I asked the price of E-6 slide processing, which the handbook never mentions. It said the handbook "does not list any charges" for E-6, pointed at the C-41 price table, and then pushed one sentence too far by concluding the lab "does not offer" it. Not listed isn't the same as not offered. But no made-up price, which is the failure I actually worry about with these tools. A follow-up about booking all 9 hours in one go kept the context and correctly hit the 3-hour cap.

Free gives you 20 messages and 2 chats a day, with a "Resets at 9:00 PM" that never says whose 9 PM. Everything ran on the Fast model; clicking Quality just opens the Plus window ($14.99 a month, or $7.58 a month yearly). The longer story is in my ChatPDF review (4/5).

A checkbox where the chat should be

Humata got a one-page note on attention mechanisms that I'd made myself. The verification mail arrived in about five seconds, I was on the $0 plan without a card, and the file was indexed in about 18 seconds.

Then I clicked the file row and nothing happened, except a checkbox ticking. Chat lives behind a separate "Ask" link on the right. Open the chat URL before indexing finishes and you get "Uh-oh. This file failed to load" instead of a spinner.

Once it was open, though, it gave me the nicest citation click of the lot. I asked for the attention formula and got softmax(QKᵀ/√dₖ)V in about two seconds; clicking [1] slid the PDF pane to the formula and painted it yellow. The exact-match figure (98.4%) took about 2.5 seconds, and a follow-up comparing it with the LSTM under the same budget pulled 98.4% vs 71.2% without me restating anything.

What I can't tell you is how it handles a trap or a fact split across pages, because my note was one page and I never asked it something that wasn't there. The free ceiling is also muddled: 60 pages, and either 10 answers (upgrade window) or 100 questions (settings page). OCR for scans is a Team-plan thing at $49 per user a month. Details in the Humata review (4/5).

Dragging a cursor through someone else's abstract

Explainpaper doesn't really answer questions about a file. You highlight a passage and it explains that passage in the margin. I used its built-in copy of "Attention Is All You Need" rather than my own upload. The dashboard's sample row didn't open on the first click and my first couple of drags didn't select anything, which is a slightly silly way to start a reading tool. Once the highlight took, the explanation came in about five seconds at the Undergrad level: a real paraphrase, no citation, no numbers. The follow-up in the margin, about what dropping recurrence means for training, also took about five seconds and went to parallel processing and vanishing gradients. No counter, no wall, and the free plan says "Free for everyone, forever." Pro is $16 a month for things like whole-paper summaries. My Explainpaper review (4/5) has the rest.

Four hundred good words under a card

SciSummary is the one I'd have liked most if I'd been allowed to read it. I gave it a two-page agronomy paper I wrote, with numbers I knew cold: moisture 18.2% to 24.7%, yield up 21%, evening versus morning drip at p = 0.41, hail in block 7.

The summary took about 43 seconds. The first text in the slot had the numbers right but padded the limitations with generic ones my paper never mentions. A little later the same slot held a 403-word version that got everything, down to p = 0.38 for mulch alone and the warning about plastic mulch and sandy soil. Careful work.

On screen I could read about one sentence of it. The rest sits under an "Activate your trial to access this summary" card that doesn't close, in a box that doesn't scroll. A second summary was refused outright, chat and figures are locked, and the trial goes through a checkout that wants a payment method before day one. I backed out. Hence the SciSummary review (3/5).

Which tab I'd actually leave open

For a handbook, a lease or a course reader where I need one rule and the line it came from, I'd open ChatPDF first. It's quick, it works before you even sign up, and it said "not listed" when the answer wasn't there. I'd still click every chip on an answer that spans pages, because I've seen it skip one. If the PDF is a clean, short research note and the click-to-highlight loop is the whole point, Humata did that best, provided you find the Ask link and stay under its small free allowance. If I'm reading a paper and keep tripping on sentences rather than hunting for facts, Explainpaper is the one I'd keep beside it, since it never put a meter or a paywall in front of me. SciSummary only makes sense if you were already going to pay; its summarizer is good, and its free plan is a preview of a summary rather than a summary.

If you're down to two, the pairwise pages are here: ChatPDF vs Humata, ChatPDF vs Explainpaper and Explainpaper vs Humata.

Sources: https://toolifynav.com/item/chatpdf-vs-humata-vs-explainpaper-vs-scisummary (official website, verified during review).

I opened felo.ai on October 14, 2026 (ART) with one job: ask a three-part fact check that includes a lie, see whether the citations open, then try to turn the answer into something I can keep. I did not need an account for the search half. The keep half stopped me cold.

The homepage before I typed anything

The tab title calls it a free multilingual AI search and creation platform. The hero on my visit rotated through "All docs. Together. Evolving." A promo modal pushed text-to-video ("One sentence or one image. Your next video.") and I dismissed it with Try Later. No cookie banner showed up.

The model strip waved GPT-6 Astra, GPT-6.1 Sol, Claude Sonnet 5.5, and Claude 5.5 Opus. Feature chips under the ask box: Text to Video, AI Slide (Hot), AI Designer, Image to Video, Research. It already looks like a workspace, not a thin answer box.

One anonymous query with a planted trap

I stayed logged out on purpose and asked:

> What is the capital of Argentina, and does UNESCO list Iguazu Falls as a World Heritage Site? Also, is Mount Aconcagua located in Chile?

Ground truth I already knew: Buenos Aires; Iguazú is UNESCO; Aconcagua is in Argentina, not Chile. That Chile claim is the trap.

From send to a finished answer: about 21 seconds. The lead paragraph got all three right, and put not in bold on the Aconcagua line. Sections underneath covered Buenos Aires as capital since 1880, Iguazú National Park inscribed in 1984 and the Brazilian side in 1986, then Mendoza province for Aconcagua at 6,959 m and roughly 15 km from the Chilean border. The UI showed 30 Source & Thinking.

I opened two citation chips. Wikipedia Buenos Aires loaded. Britannica Buenos Aires loaded. The UNESCO World Heritage URL hit a Cloudflare "Just a moment..." interstitial in this browser, but the URL itself was the real WHC list page. Two of three opened cleanly; the third was blocked by Cloudflare, not by a dead link.

That is the product working. No signup wall. Real sources. Trap caught.

Rewrite it in Spanish without logging in

Still anonymous, I typed a follow-up: rewrite the answer in Spanish and keep the citations.

About 31 seconds later I had a Spanish lead paragraph that still refused the Chile claim, then Spanish sections for capital, Iguazú, and Aconcagua with numbered refs intact. Source count jumped to 41 Source & Thinking. English ask, Spanish answer, citations kept. That is the multilingual pitch doing something I could measure, not a brochure line.

Where signup stopped

I tried email signup with my Hotmail address — no Google, no Apple, no SSO. Felo treated the address as new, asked for a password, then asked for an emailed code. The message from no-reply@felo.ai landed in Junk in about 55 seconds, subject carrying the six-digit code. The on-screen countdown was only 60 seconds, so Junk plus the timer is a tight squeeze.

After I entered the code, the page threw a reCAPTCHA image grid: select all squares with crosswalks. I do not solve those in reviews. Skip produced "Verification failed." No account. No history. No free-tier credit counter I could burn.

So everything below is either anonymous UI or the public pricing page.

Pricing I could read without logging in

At felo.ai/en/pricing:

• Standard / Free forever: 200 bonus credits/day, unlimited high-speed search, 5 professional searches per day, 3 file analyses per day. Advanced models and the pro workspace (AI Slides, AI Designer, custom Agents, video) are called out as not included. No card required.
• Pro monthly: $14.99/month (strikethrough $20.99). 15,000 credits/month plus 200 bonus/day. Lists GPT-6 Astra, Claude Fable, Grok 4.6, image and video models, AI Slides, Deep Research, API.
• Pro annual: $12.50/month billed as $149.99/year (save 16%). Same feature set.

Credit footnotes on that page: Pro Search 15 credits, Research Agent 150, slide generation 30 credits/page (promo), image 50, 2K video 138 credits/second. Mindmap marked 0 credits for a limited time.

I never hit the free daily caps anonymously. The numbers are quoted from the pricing page, not from a wall I triggered.

The creation path I could not finish

On the answer thread I clicked Generate AI Slide (Upgraded). Instant modal: "Welcome to Felo — Login or register to unlock all features." Same Google / Apple / email choices. Without an account, slides, LiveDoc, file analysis, and the agent toys stay behind that door. Fair product choice. For this session it means I scored the search half hard and the creation half only as a login wall.

Who this is for, from what I saw

If you need a cited answer across languages and you are fine starting without an account, Felo cleared a bar Perplexity-class tools sometimes miss on multilingual follow-ups. Students and analysts who live between English and Spanish (or Japanese/Chinese sources, which I did not retest tonight) get the headline value on free high-speed search.

If you need slides, LiveDoc, or file analysis on day one, plan on clearing email verify plus an image captcha — or use SSO, which I refused. Academics who need peer-reviewed databases still need those databases. Teams that want on-prem are not in this product.

I did not test logged-in Deep Research, MindMap, video, API, or credit exhaustion. I am not guessing those.

Final rating: 4 out of 5. Search and multilingual rewrite were strong anonymously; signup friction and a hard login wall on AI Slides keep it from a five.

Sources: https://toolifynav.com/item/felo (official website, verified during review).

The PDF was a two-page handbook I'd written myself for a darkroom that doesn't exist, "Riverside Film Lab — Member Handbook (2026 edition)." Membership prices, booking rules, opening hours, a developer rule, a small price table for C-41, a volunteer perk. I spread the facts across both pages on purpose, because the thing I wanted to know about ChatPDF was less "can it read" and more "when it answers, can I see where from."

Hey there, film enthusiast

chatpdf.com opens on "Best-in-class AI tools for students and researchers," a row of university logos, and a drop box that says "Drop a file or upload." The FAQ is blunt about the free side: two documents a day, and no account needed to start. That held up. I dropped the handbook in without signing in, and after about 15 seconds it said "All 2 pages checked, ready to chat!"

The auto-summary greeted me with "Hey there, film enthusiast!" and five bullets in a slightly too casual voice ("You gotta book your sessions on a paper calendar…"). The bullets were accurate, just vague. Under them sat three suggested prompts, including "How can volunteering impact my membership benefits?", plus Flashcards and Slides buttons and a sidebar nudge to sign up and save the chat history.

Signing up was the least eventful part of the night. The modal has a Google button, but also a plain email and password form, so I used my Hotmail address. No captcha, no verification mail, no code, no card. I was in straight away and the anonymous chat came along with me. I checked the inbox and the junk folder later out of habit; nothing from ChatPDF in either.

The steward hours that lost their page

The first question was the developer one above, answered in two or three seconds, correct, and cited to the right page. So I made the second one harder: "I'm a student member and I volunteered 4 hours as a steward in March. How many darkroom hours do I get in April, and can I book any of them on a Monday?"

That needs both pages. The 6 hours for a student membership and the Monday closure are on page one; the rule that 4 hours of stewarding earns 3 free hours the next month is on page two. It came back in about three seconds with 6 plus 3, so 9 hours in April, and no Mondays because the lab is closed Sundays and Mondays. Right on both counts.

Then I looked at the chips. Two of them, both ↗1. The one after the volunteer sentence jumped to page one and highlighted "student membership costs $22 per month and includes 6 hours." The steward rule on page two never got a chip at all. The maths used it, the receipt left it out.

That's a small thing on a two-page file where I know where everything is. On a 60-page lease, it's exactly the citation I'd want, and I'd have had to go hunting for it myself.

No E-6 on the price list

Next I asked for something that isn't in the document: "How much does the lab charge to develop a roll of E-6 slide film?" The handbook only prices C-41, $9 for 35 mm and $11 for 120. It's the question I care about most with this kind of tool, because a plausible made-up number is worse than no answer.

ChatPDF didn't. "The handbook does not list any charges or services related to developing E-6 slide film. It only provides prices for C-41 development of 35 mm and 120 rolls…" with a ↗2 chip that jumped to the C-41 table on page two. It did add one sentence too many, "Therefore, it appears that the lab does not offer E-6 slide film development services," which isn't what the handbook says. Not listed isn't the same as not offered. Still, no invented price.

For the follow-up I went back two turns: "Can I book all 9 of those hours in one session?" It kept the 9 hours from the steward question even with the E-6 detour in between, and said no, a single booking may not exceed 3 hours, citing page one where that rule actually is.

Twenty messages, two chats, one locked toggle

The account window is short. "Free usage today," "Resets at 9:00 PM," and after my four questions, "4/20 messages" and "1/2 chats per day." One message per question, two documents a day. The reset time has no timezone next to it, so I can't tell you whose 9 PM that is.

Under the chat box there's a "Fast | Quality" switch, set to Fast. On the free plan Quality isn't really a setting. Clicking it opens the upgrade window with a red line on top: "Upgrade to Plus to access the best model." Every answer above came from the Fast side.

Plus lists unlimited PDFs and questions, 2,000 pages per PDF, 50 PDFs per folder and the better model, at $14.99 a month or $7.58 a month on the yearly option, which comes pre-selected with a "SAVE 50%" tag. I went looking for a proper pricing page and chatpdf.com/pricing is a 404, so that window is the only place the prices live.

I opened the Flashcards dialog (few around 10, normal 20 to 40, many 50+) and the Slides one, which offered "Generate Slides (7 Slides)," but didn't generate either. Folders, Word and slide files, websites, YouTube, scanned PDFs and anything long enough to get near the page limits I didn't try, and I left AI Writer, AI Detector and Research alone.

What I'd use it for is the job I gave it: a handbook, a policy, a course reader, a contract, where you need one specific rule and want to see the line it came from. A student with a reading pack, or someone checking a lease clause before a phone call, will get a lot out of two documents and twenty questions a day. If you read a stack of papers every day, the free counters won't last an afternoon, and if you need every part of a multi-page answer pinned to its source, check the chips rather than trusting them.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/chatpdf (official website, verified during review).

Getting there was quick. Paperpal is the academic writing assistant we list under writing tools on ToolifyNav, and its homepage at paperpal.com leads with "The only AI workspace you need from search to submission," a Forbes "Best AI Writing Tool" laurel, and a strip of numbers: 1,500+ journals, 1M+ papers checked before submission, 24+ years of scholarly publishing. The big button is "Sign up with Google." I went with email and a password instead, which routes you through an Editage sign-in page, because Paperpal comes from the same company and the accounts are shared. Name, email, a password, the Terms checkbox. No verification link, no code. It dropped me straight into a blank draft on edit.paperpal.com, and when I looked at the inbox and the junk folder afterwards there was nothing from them at all. No card either.

A three-step quiz sits in front of the editor: role (I said Post-Doc/Researcher; Student, Professor and Editor were also there), how I heard about it, and what I wanted to do today. I picked "Improve Writing." One oddity: paperpal.com/pricing gave me a CloudFront 403 error page, though the plans page inside the app loaded fine.

Seven slips, eleven cards

The test text was 86 words I wrote myself, a rough take on a directory submission rule, deliberately bloggy. The seven slips: "their is one condition," "on you're homepage," "the system can checks," "plans is published," "they does not need the badge," "Every listings are reviewed," and "descriptions that makes misleading claims."

The eleven suggestions came out as Grammar 6, Readability 3, Mechanics and Style 1, Vocabulary 1. None of them were blurred or tagged for Prime. On a lot of free grammar tiers, that's where the good half of the list disappears, and here it just didn't.

All seven slips were caught. "checks" went to "check" as a verb-form fix. "plans is" became "are," and "descriptions that makes" became "make," both as subject-verb agreement, with the explanation spelled out. "Every listings are" was handled as two separate cards, one for the noun ("listing") and one for the verb ("is"), which is the right way to read that mistake.

The two I'd call plain grammar got filed somewhere else. "you're homepage" showed up under Readability as "Rephrase this portion for readability," offering "your." "they does not need the badge" was also a Readability rephrase, offering "do not require." Both ended up correct. The labels were just a little vague for errors that basic.

Then three things I hadn't planted: "keep" to "maintain," an inserted "that" after "check," and deleting the comma after "payment." I accepted the their/there fix (11 dropped to 10), then hit "Accept all" on the second sentence, which cleared four cards at once and left six. That sentence now read: "You need to place the official ToolifyNav badge on your homepage and maintain a backlink to your listing, then press Verify so the system can check that the badge is live." Correct, and a notch stiffer than I'd write it for a website.

189 before I'd accepted anything

The footer under the panel had said "200/200 free uses left" when I opened the draft. The moment the check finished, it said 189/200.

So the free allowance counts suggestions generated, not fixes you take. Eleven cards, eleven edits gone, before I'd clicked a single Accept. Accepting afterwards cost nothing; it stayed at 189 through the single fix and the batch of four. That's fine on an 86-word paragraph. On a messy chapter draft with a few hundred problems, one pass could eat most of the month.

One Make Academic, four left

I highlighted the third sentence, "Paid Premium and Highlight plans is published after payment, and they does not need the badge," clicked "⚡ Paperpal AI" on the floating toolbar, and picked "Make Academic." About 2.5 seconds later:

"The Paid Premium and Highlight plans are published following payment and do not require a badge."

Both errors fixed, the meaning intact, nothing added that I hadn't said. The panel has tabs for Paraphrase, Improve Fluency, Make Academic and Simplify, a Show changes toggle, and Replace, Copy, Retry, Discard. In the corner, "5/5 free uses left today" became "4/5 free uses left today," next to an UPGRADE button. No wall, no modal. Five a day is a handful of stubborn sentences, not a section.

I left the rest of the sidebar alone this time. There's a Research & Cite panel ("Your AI Research Assistant with 300M+ scholarly publications") with Ask a Question, Cite Text and Search Papers on its own 5-a-day pool, and a Checks area that takes .docx uploads up to 25 MB and lists Plagiarism Check, AI Detection, Reference Check, Journal Fit, AI Review and human expert editing. A banner at the bottom kept offering "Download for Word." I didn't upload a file, run a plagiarism or reference report, or try the Word add-in, so I can't tell you how any of that holds up.

What $12 a month changes

The plans page inside the app, monthly: Free $0, Prime $25, Pro $59. Billed annually, Prime is $12 a month ($144 a year) and Pro is $29 ($348). On the annual view a welcome code, US2026, was already applied and showed $9.60 and $23.20.

• Free: 200 grammar edits a month, 5 AI writing actions a day, 5 research questions a day, 5 chat-with-paper files a month, a Lite plagiarism report and an AI-detection preview
• Prime: 20,000 edits a month, 500 AI writing actions a month, 10k words of plagiarism checking, 25k words of AI detection
• Pro: unlimited edits and AI writing, 30k words of plagiarism checking

The jump that matters for me is edits. Two hundred is plenty for a cover letter or an abstract. It is not a thesis.

If you're a grad student, a researcher, or writing journal English as a second language, this is a good free way to clean up short passages, and the academic pull ("keep" turning into "maintain") is probably what you want. For blog posts, product copy or anything with a casual voice, that same pull will keep sanding your sentences into a register you didn't ask for. I'd start with a paragraph like I did, watch the 200 meter, and only then decide about Prime.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/paperpal (official website, verified during review).

Getting there was quick and a little odd. The register form at quizgecko.com is name, email, password, confirm password. I submitted it at 22:17 on a Wednesday night and landed in onboarding with no verification step. Half an hour later I searched Outlook for "quizgecko" across Focused, Other and Junk and found nothing, not even a welcome mail. The onboarding is written for students: "Are you a..." Learner or Educator, a screen asking whether ADHD, Autism, Dyslexia or Dyscalculia applies to me, "What are you studying?" with Healthcare / Nursing, EMS, Law, IT / Cybersecurity and nothing near "runs a directory". Then "Choose a plan for after your trial", where I clicked "Choose free". My account came out named WorkablePascal3985.

Fifty-three seconds, a whole course

Of the ten sources on the create screen I took Text and pasted the guide. The configure page had all five boxes ticked by default: Questions, Flashcards, Study Notes, Podcast, Mind Map. Difficulty sat on Mixed, Max Questions on Auto. I left it and hit Generate at 22:19:30.

At 22:20:23 I had a "course" called "ToolifyNav Submissions and Directory Operations". Three modules of 12, 13 and 13 questions, 31 flashcards, four study notes, a podcast and a mind map. On top sat an AI-generated header image of a fake dashboard with labels like "Kari Moden", "Flooit" and "AI Toot Geolla". On a course about my own site, that was the first thing I saw.

I read all 38 against my text. No wrong answer keys, no invented rules. There's a distractor "Badge links are follow links" marked wrong, and the true/false "Premium and Highlight submissions require a homepage badge" keyed False, which are the traps I'd set for a human. My only quibble was a matching pair, "Creative visual work => Art", shaky when Image is also one of the 13.

What wore on me was volume. There are maybe 15 distinct facts in that guide, and Auto stretched them to 38. "The payment clears" gets asked as a multiple choice, a short answer and a fill-in-the-blank, then turns up again inside two matching questions. "Click Verify" turns up four times, and one matching pair just says the same thing on both sides. If I sent this to a maker as a self-check, I'd delete half of it first.

The grader noticed which mistake I made

The editor shows "12 / 200" for the module. I rewrote question #0, both the question and its answer, hit Save Changes, hard-reloaded. It stuck.

Practice mode shuffles the options and hands out "+15 XP" per hit. Short answers are graded by the AI, so I wrote one sloppy on purpose: "make an account with a normal email, do a real task and measure the result" for what a reviewer does before scoring. Correct, with "Spot on! You nailed the process completely by mentioning signing up with a normal email, running a task, and measuring the result." Then "26 categories" for the category count, which got "Oh so close, but you doubled the number!" It saw what I'd done wrong, not just that the string didn't match.

Podcast, flashcards, and a map with five categories missing

The extras are where the accuracy thins out. The flashcards are fine: 31 of them, a 1–4 spaced-repetition rating, memory tips like "Talk, then code." for Chatbots and Coding. The podcast is an 8:32 MP3 with one narrator. It gets the rules right and then says them again. Eight and a half minutes from 273 words, with lines like "A compact memory contrast is, free means badge plus backlink, then verify".

The mind map is the weak one. Under Core Categories it made up "Creative Categories" and "Productivity Categories", fit eight of my 13 categories into them, and dropped Chatbots, Education, Image, Marketing and Social. It also filed "Nofollow Badge Links" under a branch about rejections, which isn't where it belongs. The quiz knew there were 13. The map didn't.

Free exports that work, one CSV that doesn't

The Export Questions dialog lists 19 formats, from Text, CSV, Excel and JSON to PDF worksheets, Aiken (Moodle), Canvas, Blackboard, Schoology, Sakai and two QTI versions. None show a lock. The plain ones really are free: TXT at 14,996 bytes, CSV 12,491, Excel 8,993, JSON 87,805, a nine-page PDF worksheet at 186,266 bytes and a PDF answer key at 274,158. No watermark. The worksheet's matching section wraps badly, "1. Final" on one line and "rating" on the next, but it prints.

Then I opened the CSV. Every untouched question has a letter in CorrectAnswers. Question #0, the one I'd edited, has '@. The JSON export shows why: generated questions number their answers 1 to 4, my edited one got saved as 0 to 3, and one character before "A" is "@". I unticked and re-ticked the correct answer, saved, re-exported: still '@. Edit a question, move the CSV into another tool, and that question has no valid answer.

Canvas and QTI 2.1 fail differently. Click, toast: "Export started - check your downloads". Then a new tab with raw JSON, a 403 reading {"message":"LMS exports are only available on the Ultra and Organizational plans."}. Fair rule. It should be a lock icon in the dialog.

One generate, one month

Afterwards I went back to the create page to feed it my published Napkin AI review by URL. "Course limit reached. You have reached your monthly course creation limit." Behind it, "Upgrade to continue" with "Start Free Trial". That Auto-sized, everything-ticked first run was the whole free month, so URL, PDF and YouTube input stayed untried, along with the tutor chat, what a share link looks like to a logged-out visitor, and whether the LMS files import cleanly once paid.

The plans page puts Free at 1 AI course and 1 podcast a month, up to 10k tokens per course, "Basic AI model for shorter files". Plus is $16 a month, or $9 billed yearly ($108), for 20 courses and 75k tokens. Ultra is $29, or $23 yearly ($276), for unlimited courses, 150k tokens, up to five files and the LMS exports. A "40% off" button sits in the top bar the whole time.

I'd point a teacher or trainer at it, or a founder who wants to find out whether their FAQ says what they think it says, one document a month. If you need a stack of quizzes on free, or want them in Canvas without Ultra, it runs out fast. The listing is on ToolifyNav.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/quizgecko (official website, verified during review).

That was the part I didn't see coming. I went in expecting to argue with the AI about logic and fix the diagram by hand. The logic turned out to be the easy bit.

Eight seconds for a paragraph I keep retyping

I explain ToolifyNav's submit flow constantly, to makers in email and in X replies, and it's always a wall of text. So the test input was my own paragraph: open the Submit page, pick a plan; on Free, paste the official badge and a backlink on your homepage and click Verify; if it's found you go live, if not you get an email saying what's missing and can try again; Premium or Highlight pays through checkout and goes live when payment clears.

Signup on napkin.ai was a plain email and password form, no SSO. I submitted at 22:12 and the "Your one-time verification code" mail from contact@napkin.ai showed up about 15 seconds later, in Outlook's Junk folder, not the inbox. Code in at 22:16, then a five-question survey (For work, Just me, Engineering, Blogs and newsletters, how did you hear about us), then "Thanks Leo! You're all set to create your first Napkin!" No card at any point.

A new napkin offers "By pasting my text", "By describing my idea" or "By importing a file". I pasted, clicked the small blue spark next to the paragraph at 22:19:48, and the AI Suggestions panel had six thumbnails at 22:19:56.

The one it showed first was a mind map with bubbles like Submit Page, Verification, Plans and Listing Live. Pretty, and useless here: the whole point of the paragraph is the if/else, and the mind map threw the order away. The vertical flowchart was a different story. "ToolifyNav Listing Process" starts at Submit Page, hits a "Select Plan" hexagon that splits into Free and Premium/Highlight, runs the Free side through Verify Listing into "Requirements Found?", sends No to "Send Missing Email" with a line back to Verify, sends Yes to "Listing Goes Live", and joins the paid side there after "Payment Clears". It added a few subtitles I didn't write, like "Wait for payment confirmation". Those were fine.

The wiring, and the rename that renamed the wrong thing

Right logic, sloppy routing. The retry loop drops under "Listing Goes Live", runs all the way up the right edge of the canvas, and crosses the main line on the way. The payment arrow and the Yes arrow merge into one entry. You can read it, but I wouldn't put it on the Submit page untouched.

So I started editing, and the first edit went somewhere else. I tried to change the top node from "Submit Page" to "Visit Submit Page" and the text landed in the document title above the canvas. The napkin is now called "Visit Submit Page", and every export after that downloaded as "Visit Submit Page - Visual 1.png". The node itself only changed when I double-clicked right on its text, which opens an inline editor with a tiny toolbar (Roboto, 12, B, a color picker). The box widened to fit and nothing overflowed.

Color is where I lost the most time. I wanted one box orange. The Colors panel doesn't do one box; it has Background and a row of Elements swatches, and each swatch is a role, so every process box went orange and cream at once. Somewhere in that, one swatch ended up white, and that role also drives the connector labels and outlines. Nothing in the panel tells you which swatch owns what. Some of that is on my clicking. The rest is on a default transparent background that makes white text vanish without a trace.

Swap Layout was greyed out on this flowchart. The Effects panel has styles like Ink Print, Blueprint Review, Sticky Notes and Origami, each behind an "Apply 100" button, which is 100 credits, so I left them alone. Undo worked every time I needed it.

Export sizes, the watermark, and where the credits hide

The default 1x PNG came out at 542 × 1665 px, 151,995 bytes. That's too narrow for a blog column, and I almost wrote it off before finding Resolution in the export options. 2x gave me 1084 × 3329 px (375,629 bytes) and 3x 1626 × 4994 px (642,964 bytes), both free. Every PNG carries a small grey "Made with Napkin" in the bottom-right corner, and flipping Napkin Logo to Off gets you "Remove Napkin Logo is a premium feature." PNG and PDF are free; SVG, PowerPoint and Google Slides each show a 3-credit badge in the Export Visual dialog. I didn't try those, or the PDF, or decks, file import, Brand Studio, sharing or the Developer tab.

The second test was a short paragraph about how I plan a week of writing, four steps in plain sentences. Four suggestions came back in about 14 seconds (22:27:09 to 22:27:23). Each box had the right sentence, but in the order I typed them, so the last step sat under the wrong day like it belonged to the next week. A human would have caught that. Then again, I wrote it in that order.

The editor never shows your balance. I found it in Account Settings, under Subscription: "Weekly Credits, 466 Credits left", refilling October 11, 2026. Two generations cost 34 of the 500. On napkin.ai/pricing, Plus is $9 per person per month billed annually ($12 monthly) with 10,000 credits, SVG and PPT export and no branding; Pro is $22 ($30 monthly) with 30,000 credits, custom branding and font upload.

Before it goes anywhere near my Submit page

If you already write your processes out in prose, for docs, a changelog, or a long X reply, and you want a real flowchart out of that paragraph without opening Figma, this does the hard part. It found my branches and my retry loop on the first try, and the free tier gives you enough credits and a big enough PNG to use the result. Just put the export on the page you're publishing to before you trust it.

If you need to style single boxes, route lines yourself, or ship an image without someone else's logo on it, the free plan will annoy you fast. Pricing and alternatives are on the Napkin AI listing.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/napkin-ai (official website, verified during review).

The file was a 512 × 512 JPEG, 90,406 bytes, a shrunk copy of the rainy Buenos Aires street I'd made in NightCafe the night before: neon signs, wet cobblestones, one guy on a bike riding at the camera. I made the small copy on purpose, because I still had the real 1,024-pixel original (631,294 bytes) and wanted a control. An upscaler should get close to that original. When it beats it, something is being invented.

Ten credits and a 512 px file

Signup on letsenhance.io was the plain kind. Email, password, an opt-in box for "fresh news, discounts, and updates" that I left alone, "Create an account". I ignored the SSO buttons. That was 22:16. The "Please validate your email" mail was in my Outlook Focused inbox within a minute, not Junk for once, and "Confirm my email" bounced me through a "Thanks, your email was confirmed!" page into the app at /boost. No survey, no card. The header said "10 credits left", next to a green "Invite friends!" button.

The upload area takes "JPG, PNG, WebP up to 50 mb", from disk or Google Drive. Once the file landed, the row did the math I'd normally do in my head: input "512 × 512 px · 0.3 MP", output "2048 × 2048 px · 4.2 MP", and a print size of 6.83 × 6.83″ at 300 DPI. So, a little under seven inches square. Not a poster, but a real print, which is all I wanted to know.

On the right, "Upscale & Enhance" was switched on with Prime selected by default, next to Gentle, Old photo, Try all and More models. Scale went to 4x. 8x and 16x were there but locked on the free plan. I left Tone enhance and Color enhance off and kept JPEG output. One image costs one credit.

I hit Enhance at 22:27:18. The status said "Processing... Will be ready in about a few minutes Prime x4". It was ready at 22:27:51, about 29 seconds later, and the counter went from 10 to 9. "About a few minutes" for a half-minute job is an odd thing to promise, but I'll take that over the reverse.

Prime, the loupe, and a man I didn't make

The result opens in a modal called "Upscale Prime x4": a split slider over the whole frame, plus the zoom loupe underneath. The loupe is the most useful thing in the product. You drag it to the place you're worried about and see 512 against 2048 at the same spot, no exporting to another app.

The Prime download was buenos-aires-cyclist-512_LE_upscale_prime_x4.jpg, 2048 × 2048, 1,645,636 bytes. I cropped the same patches from it, from a plain bicubic 4x of the 512 file, and from the real 1024 original scaled up. On the cobblestones Prime is impressive: the reflections turn into hard little specular highlights that read as wet stone. The car headlights pick up a slight glow ring. The neon signs were gibberish to begin with (one vertical sign spells something like "EAOSOA") and they stay gibberish, just with cleaner edges. Then there's the face, sharper than the 1024 original ever was. Plausible, and wrong.

Then I found the watermark, bottom-left, sitting on the wet stones: a semi-transparent circle logo and "Let's Enhance .io", roughly 300 pixels wide on a 2048 frame, about a seventh of the width. It's on the My Images thumbnails too. That is the one thing a print-prep tool shouldn't do to a trial file, because the whole point of the trial is to print one and look at it.

Gentle changed my mind

I'd gone in assuming Prime was the obvious choice. The homepage even has a pill for it, "NEW Prime: sharper photos, real texture". For the second run I uploaded the same 512 file again and picked Gentle, still 4x, toggles off.

It finished in about 22 seconds, 22:29:15 to 22:29:41, and took me to "8 credits left". The download came back as buenos-aires-cyclist-512_LE_ml_resize_x4.jpg, 1,155,509 bytes. Not "gentle" in the name, "ml_resize". Small thing, but if you run a few models on the same file you'll be renaming downloads by hand.

Side by side, Gentle is softer. The rain streaks are faint lines instead of white needles, the edges have less of that sharpened outline, and the cyclist keeps the face he had in the 1024 original, give or take some blur. Against my control it's simply closer. Prime wins the first glance. Gentle is the one I'd send to a printer for a photo I care about, especially with a person in it. For product shots or illustration, Prime's punch probably matters more than faithfulness.

Paying to lose the logo

The pricing page, "Let us enhance your images", has three tabs. Subscriptions: Starter is 100 credits a month at $9 billed annually ($12 monthly), Pro 300 credits at $24 ($32 monthly), Max 500 credits at $34 ($45 monthly), with unused credits rolling over up to six months' worth and output ceilings of 256, 350 and 512 MP. Pay as you go is 20 credits for $9, 50 for $19, 120 for $39, and those never expire. The business tier starts at $72 a month for 1,000 credits and is the one that spells out "no watermark".

For someone who needs a handful of print files a year, the $9 pack is the sensible door, about 45 cents an image. Ten free credits mean ten images; I used two.

There's a lot more in the app than I touched: the Generator, Chat Editor, Image to Video, the new Video Upscaler, the Animate button on the result, the Old photo mode, Try all, the Balanced/Ultra/Strong/DigiArt models and batch upload. I didn't try any of them, or anything above 4x.

It makes sense for photographers and print or merch sellers who have a small file and need to know whether it survives 300 DPI, and who are fine paying a few dollars once it does. The free tier answers that question well, thanks to the loupe and the 4.2 MP preview math. It won't hand you a clean file, though. And if there's a face in the frame, switch off the default and use Gentle. The full listing is on Let's Enhance on ToolifyNav.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/lets-enhance (official website, verified during review).

That was NightCafe Creator, at creator.nightcafe.studio, which we list under image tools on ToolifyNav. The homepage says "The Home of AI Art" and puts a prompt box under it with "No payment or credit card required" in small grey type. A red banner across the top says "Save 50% on Annual Plans — Claim 6 months FREE." That banner follows you onto every page for the rest of the session.

Getting to the first free frame

Going to the studio logged out gives you "You need to be signed in to do this," with two choices: "Create an instant temporary account" or "Login now." I didn't try the temporary account. Login offered Google, Apple, Facebook and Email; I took Email, typed my Hotmail address and a password, hit Create Account at 22:23:26, and got the "Verify Your Email" screen. The mail — "Verify your email for NightCafe Creator" — was already sitting in the Outlook Focused inbox, not Junk. One click, "Your email has been verified," Continue.

There's still a little gauntlet after that. A terms screen with a required checkbox, plus an optional marketing-email box labeled "EARN 3 CREDITS" (yes, they pay you to accept newsletters). A Cloudflare Turnstile check that passed on its own. Then two survey modals asking what I plan to use it for and how much experience I have; I skipped both. From signup click to studio, about a minute.

The top bar showed 5 credits next to an orange "Offer!" pill with a 20 badge on it.

The Create panel opens in Advanced mode. The default model is Flux 2 Klein 9B Fast, with tiles for GPT Image 2 Low and Muse Image next to it, and right under the picker a line that says "Get unlimited Flux 2 Klein 9B Fast on the AI Hobbyist plan →". One image per run by default (1, 4 or 9), aspect ratio 1:1. I'd written "35mm film photo" in the prompt and still got a square, because I didn't change it — there are eleven ratios including 3:2 and 16:9, and I should have picked 3:2. Prompt Magic is switched on by default. Bottom of the panel: "Your first creation of the day is free!" and a pink button reading Create FREE.

The prompt was rainy neon street in Buenos Aires at night, 35mm film photo, reflections on wet cobblestones, a lone cyclist. Balance before: 5. After: 5.

Evolve, and a warning about my own settings

The creation page has a long row of actions: Animate, Prompt to Edit, Upscale, Creative Upscale, Expand, Adjust & Crop, Fix Faces, Remove Background, Duplicate, Evolve, and Publish, which advertises "EARN 1 CREDIT." Upscale lists Real-ESRGAN x4+, x2+ and a few variants, at 2 credits. With 5 in the account I left it alone.

I used Evolve instead, which drops the previous image in as the start image, and added , golden hour neon glow, dramatic atmosphere to the prompt. The panel now said "Goal: Create 5 Images 2/5" and "19 applicable freebies. View freebies." So the 20 on that Offer badge was a freebie counter, and one had been used. Still Create FREE.

Clicking it threw a modal: "Warning: Prompt To Edit Model with Prompt Magic ON - You're using prompt magic while using a start image with a Prompt to Edit model. Your prompt will be edited and may not work as expected." A default setting warning me about another default setting. I confirmed at 22:28:07; done at 22:28:15, about 8.5 seconds. Balance still 5.

The second frame kept the composition almost exactly: same cyclist, same lean, same buildings, same cars. What changed was the light, which went heavy amber and copper with visible rain streaks. More golden than neon — the teal mostly disappeared. If you're trying to grade a scene a different way, that's the useful behaviour. If you wanted the neon pushed harder, Prompt Magic seems to have read "golden hour" as the main idea.

More → Download gave me FEGsl4imrMU9SZip5Hdb--0--jzML_.jpg: a 1024 × 1024 JPEG, 631,294 bytes, no watermark, no card prompt. Fine for a moodboard or a social post. Small for anything you'd crop hard or print.

Five credits, nineteen freebies, and the pricing page

This is the part I had to piece together. On a free account there are credits (I had 5), freebies (19 left after two runs), and the plans talk about "600+ relax credits/mo" on top of fast credits. Nothing on the free side explains all three in one place. My two runs cost zero credits; the first thing that would have cost real credits was the upscale. I didn't wait a day to see whether the freebies refill.

/pricing, monthly:

• AI Hobbyist $12.49 — 200 Fast Credits/mo
• AI Enthusiast $24.99 — 500 ("POPULAR")
• AI Creator $39.99 — 850 ("BEST VALUE")
• AI Artist $62.49 — 1,400 ("POWER USER")

Annual billing ("SAVE 50%") brings those to $6.24, $12.49, $19.99 and $31.24 a month; quarterly is "SAVE 15%." One-off packs run from $7.99 for 100 credits to $299.99 for 10,000, with 500 at $19.99 in between. I didn't open checkout.

Count the upgrade prompts I saw in ten minutes: the red banner, the Offer pill, the Hobbyist line under the model picker, a "LIMITED TIME OFFER — Go Annual, Get 6 Months Free" card on pricing, and the credit rewards for newsletters and publishing. None of them blocked anything. They're just always in the frame.

Where it fits for me

I'd use this for what I used it for tonight: blocking out a lighting idea before a shoot, or making a quick reference frame for a client who can't picture "wet street, warm practicals." Students and hobbyists poking at AI art for the first time can get real images out without paying, and the community side (challenges, the Promptle daily game) is clearly aimed at them; I didn't touch either. It's the wrong tool if you need readable signage in the image, large native files without paying for upscales, or a quiet interface.

Things I didn't get to: video through Animate, any upscale, Fix Faces, Remove Background, Expand, the temporary account, the mobile app, and the license terms for commercial use. Next time I'd switch Prompt Magic off before evolving and set 3:2 from the start.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/nightcafe (official website, verified during review).

The train screen is blunt about what you’re doing. "Choose 50 colors. These colors will be used to train a color generator algorthim personalized to you…" — yes, "algorthim," that’s in the live copy. A note under it says for this beta your data stays in localstorage, only this browser on this device. Swatches carry labels like RICH LIGHT BLUE and MUTED MIDTONE VIOLET. When the counter hits zero, a green Start Training button appears and a navy "READY TO TRAIN" card asks you to hit Got it. Still no account. Still no card. The ToolifyNav listing for Khroma already called it a free AI color tool; I wanted to see whether that held once the neural net actually ran.

The fifty-swatch tax, then a 48-second run

Fifty clicks is a real onboarding tax. You can’t mash random tiles if you care what comes out later — I leaned dusty pastels, deep navy, warm earth, the kinds of hues I actually keep in photo grades and UI mocks. Then Start Training. Progress text in the browser, not a spinner to a remote GPU: "Training: 62% Accuracy: 95.06%" flipped into "Training: 67% Accuracy: 95.05%" and finished at 100% in about 48 seconds. Redirect to /generator. Under the hood it’s a client-side synaptic.js Perceptron writing weights into IndexedDB via localforage — which is why there’s no signup gate and also why wiping the browser means wiping the model unless you export.

The generator is an infinite scroll of pairs. Top bar has view icons for Type, Poster, Gradient, Image, four-color Palette, and Favorites. Two Bias fields both sat at 50. Search placeholder: "What are you looking for? Search 'tips' for help." A small "Generating…" bar flickered at the bottom while more cards streamed in. Heart on a card dumps it into state.favorites with the current weights. I checked that with a quick IndexedDB peek — favorites array grew, weights stayed. No toast asking me to upgrade. No credit meter. BETA badge still red next to the K logo the whole time.

"pastel" returned twenty-four pairs

I typed pastel into the search field and hit Enter. Grid snapped to 24 filtered combinations — Deco & Chambray, Royal Blue & Sweet Corn, Caper & Elm among them. Names come from ntc.js, so you’re not stuck staring at naked hexes in the feed. Poster and Type views put the same pairs into layout and body-copy samples; Gradient mode was the cleanest cover shot of the night.

Opened Color Details on Surf & Persian Blue:

• Surf — #B6D8BB, rgb(182,216,187), bias 75%
• Persian Blue — #242EBB, rgb(36,46,187), bias 100%
• WCAG — AA (6.2:1)

One click on #B6D8BB and a toast said "Copied to clipboard!" That was the whole export path for day-to-day use. No ASE gate. No "Pro unlocks hex." Just the hex, RGB, name, bias, and contrast ratio on the card. I favorited the pair before moving on — heart SVG, no confirmation modal.

Settings, JSON suitcase, one DOM quirk

Gear icon opens Settings. Chromatic Gradients was ON — "This will make some gradients look less muddy." Never fail WCAG was OFF — "This will block type combos which fail the WCAG contrast ratio test." There’s an Edit training colors link that warns it resets the generator. And a mint Download Khroma data button that packages your personal generator and favorites as JSON so you can upload them on another machine’s training page. That’s the migration story: not cloud sync, a file. I didn’t open Figma plugins or try this on a phone — stayed in the desktop browser session from about 22:19 to 22:26 ART.

One bug showed up after a few filter passes. Search replaces #grid children; if Color Details was open on a card that got torn out, the overlay detaches until you reload. Minor, but real. Unicorn promo cards from the same creator sit in the feed with a "Start now free" pitch — I left those alone and stayed on Khroma itself.

Khroma is for designers and indie builders who will trade fifty deliberate likes for a taste-trained feed and one-click hex/WCAG without making an account. Skip it if you need cloud sync out of the box or you refuse to pick that many swatches before you see a single palette. Free means free here — no tier table, no card wall, local weights, JSON suitcase if you move machines.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/khroma (official website, verified during review).

The file was a 53.5 KB watch product shot I’d marked up on purpose — red SAMPLE WATERMARK bottom-right, blue FOR SALE — DEMO top-left, dark REMOVE ME across the crystal. Homepage copy still sitting in the tab behind the modal: “Remove any unwanted object, defect, people or text from your pictures in seconds.” Upload box: “Click here or drag an image file.” That’s the whole pitch. Tuesday night in La Rioja, ~22:26 ART, I just wanted to know if the free brush could wipe those three overlays without a billing form in the way. The listing on ToolifyNav already said freemium with a ~720p free export; I wanted the live modal numbers, not the marketing blur.

Three strokes on a watch photo

Brush default sat at 49.9 on a 10–200 slider. Lime-green eraser active. Nothing else competing for attention — undo, zoom, gear, a bright Download that stayed quiet until the first fill landed. The canvas itself was already the downscaled frame. You feel that before you paint: the watch is a little softer than the JPG I dropped in.

First pass: painted over the red watermark. Release → about 2.0 seconds → gray tabletop filled back in. No smear edge where the letters had been. Download lit up. Undo was there if I wanted it.

Second pass: blue FOR SALE — DEMO. Roughly 1.8 seconds. Canvas updated in place. No interstitial. No credit meter. No “one more free erase” toast. That silence is part of why this tool feels usable — the free path doesn’t nickel-and-dime each stroke.

Third pass was the one that usually ruins product shots — dark text sitting on glass. Tight brush over REMOVE ME, and the crystal highlights came back clean enough that I’d actually ship the crop for a web thumbnail. Three strokes, maybe ten seconds of hand work, about six seconds of server time stacked. I didn’t open ClipDrop’s other tools, the API, or mobile gestures. Tonight was the Cleanup canvas only.

The download is honest about what free means

Green Download. Instant. File landed as test-clutter_cleanup.jpg, 196,721 bytes (~192 KB). Resolution check: exactly 720 × 523. The payload looked like PNG bits wearing a .jpg name. Zero ads. Zero captcha. Zero card wall.

That 720p wall is the whole free story. My upload started at 1200×872 and lost more than a third of its pixels before I ever touched the brush — the HD modal forced the choice up front, which I prefer to a surprise watermark on export. Fine for messaging a cleaned mockup or a small ecommerce thumb. Not fine if a client expects the original megapixels back. The upgrade modal later spelled it out again: “Upgrade now to enable high resolution with no compression (vs 720p max for the free version).”

I bounced between the eye toggle and the filled canvas once after the third stroke, just to confirm the overlays were actually gone and not merely dimmed. They were gone. Undo still worked after the download button appeared, which matters when you’re mid-session and brush a little too wide on a dial marker.

Magic link went to Junk, pricing didn’t hide

I still signed in after the download, mostly to see how accounts work. Menu → Sign in. Passwordless. Typed my Hotmail address. The inline copy said a sign-in link was on its way and added “(check your spams folder).” They weren’t kidding — Outlook Junk, sender noreply@cleanup.pictures, subject along the lines of “Sign in to cleanup-pictures requested at 2026 September 30 01:29 UTC.” Firebase link. Session flipped to free with Sign out visible. No Google / Apple / GitHub / Microsoft buttons in the path I used.

What the site actually showed for money:

• Free ($0): unlimited images, hard-capped at 720p
• Cleanup Pro: $5/month or $36/year ($3/mo billed annually, marked down from $60), 3-day trial, unlimited resolution + high-quality refiner
• ClipDrop Pro: from $11/month if you want the wider suite (remove/replace background, uncrop, generative fill)

API is listed as usage-based. I didn’t poke it. Didn’t run the paid trial checkout either — the “Try Pro 3 days free” button was right there on the HD gate, and I left it alone so the free path stayed clean. Same ecosystem as Clipdrop (the site brands itself “Cleanup.pictures by Clipdrop”), so if you already live in that suite the Pro upsell is familiar territory. I stayed on Cleanup alone.

For marketplace sellers, real-estate people, and anyone scrubbing date stamps or sample text off an otherwise good photo, this is a fast browser eraser. Drop file, paint, download. Anonymous works. Magic link works if you dig Junk. The catch is resolution: if your pipeline needs full-res client files, free will bounce you at the HD modal every time. Pay $5 or take the annual $36 when that wall actually blocks a deliverable — not before.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/cleanup-pictures (official website, verified during review).

Suggestions landed in about 1.5 seconds after the paste hit https://languagetool.org/editor/welcome. Ten cards stacked in the Correct tab. The first one was blunt: Their → There, labeled “Use the right word,” with a blue replacement button. I pressed it. The word swapped inline in the canvas. The score jumped to 76. No refresh. No card form. No “start trial” toast in the way of a basic fix.

I’d already signed up a few minutes earlier with email only — more on that below — so the editor knew me as Leo Liang. A sticky banner still said my Hotmail address needed confirming. I kept writing anyway.

What the free checker caught on that draft

Default mode, Picky off, cleaned up a lot of the planted junk. alot → a lot under “Correct the spelling error.” actualy → actually. tryed → tried. recommand → recommend (the card said “Change the verb tense,” which is a funny label for a misspelling, but the fix was right). english → English. The results was → were for subject-verb. write ups → write-ups. doesnt → don't under “Fix the auxiliary or modal verb.” Even the em-dash spacing: trust — with → trust—with.

It also walked past some things I expected a grammar tool to poke on a free pass. an other (as in “an other one”). to much in “spend to much time.” Inside quotes, Its a good idea sat quiet. Me and my cofounder didn’t get a look either on that first basic pass.

There’s a Picky Mode toggle in the sidebar. Flip it and the issue count went from 10 to 16 — smart quotes and other typography noise showed up in the list. After Picky was on and after the There fix, the counter sat at 11, and the editor did highlight Its and Me and my in the text. to much and an other still didn’t move. That mix is the free tier in one sitting: fast on the obvious spelling and agreement misses, uneven on a few basics until you turn the picky dial, and silent on a couple of planted traps.

I didn’t poke the Chrome or Firefox extensions in Gmail or Google Docs, the Word / LibreOffice / Apple Pages add-ins, the desktop apps, team style guides, non-English Premium rules on a long file, or the API. This session stayed in the browser editor.

Three Standard paraphrases, then the Premium tones

The separate paraphrasing page at /paraphrasing-tool is honest about the free ration: three AI sentence paraphrases per day, Standard tone only. I fed it “ToolifyNav helps people find the right AI tool faster.” Standard came back with “Finding the ideal AI tool is made easier with ToolifyNav.” Fine. Usable. A little flatter than what I typed, which is what I want from a paraphrase when I’m stuck, not a marketing rewrite.

Fluent, Humanize, Formal, and Academic each opened the same gate. Modal copy, roughly: Formality Mode / Premium Feature / Unlock Premium Modes / Unlimited rephrasings / Advanced style suggestions. So you can taste a rewrite on free. The tone dial is paid. If your whole workflow is “make this sound more formal,” you’ll feel that wall on day one.

Email in, no card — then the 2,000-character wall

Signup was email-only on languagetool.org/register. Sign up with e-mail, name Leo Liang, my Hotmail address, password, single-click reCAPTCHA — just the checkbox, no image grid. No Google, Apple, Microsoft, or GitHub. Terms modal, skip the survey at /survey, plan screen titled “Sounds Like a Good Plan,” then Continue For Free. Editor opened. The banner stayed sticky: Your e-mail … still needs to be confirmed. Resend e-mail. I checked Hotmail Inbox and Junk; I never needed the confirm link to keep pasting and accepting fixes.

The free cap is 2,000 characters per text field. You feel it before you paste your own stuff. The default welcome sample in the editor was already 2280/2000 Characters — orange counter, red “Text too long” where the Text Score usually sits, Premium pitch up to 150,000. That alone is the free tier’s loudest honesty. My 814-character draft fit. A medium blog section would not.

Premium pricing from the UI I opened that night:

• 2 Years: $119.80 ($4.99/mo)
• Annual (Bestseller): $69.90/yr ($5.83/mo)
• Quarterly: $49.90/q ($16.63/mo)
• Monthly: $24.90/mo
• Teams: from $125.80/yr for two seats

Premium also unlocks the longer field, unlimited paraphrasing across those tones, advanced style, and the office plugins I didn’t install. Free keeps basic grammar, spelling, punctuation, limited Picky, and those three Standard paraphrases.

Short emails, blog intros, UI copy snippets — the free web check is quick and doesn’t ask for a card. Long paste jobs and the fancy paraphrase tones hit the wall early. I left Monday night with a cleaner draft, a score that moved when I clicked, and a clear map of what free will not do.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/languagetool (official website, verified during review).

I wanted a first draft I could edit tonight — not a marketing site, not a paid trial that asks for a card before you see a slide. Email path only. My Hotmail address, Continue with email, no Google/Apple/Microsoft. Verification mail from Gamma ("Please verify your email address") hit Hotmail Inbox in roughly fifteen seconds. Setup: Leo / Liang, password, then the onboarding taps — For work, Creator, Just me, came from search. Dashboard said Leo Liang's Workspace Free, 400 AI credits, Gamma 5 beta banner up top. No card form.

First night in Gamma 5

The welcome modal, a tiny tour, and a feedback nudge stacked on each other the first time through. Three dismisses before the canvas felt mine. Minor, but I muttered at it. Gamma 5 still wears a purple "currently in beta" chip; the create card offered Presentation, Document, Carousel, Image, Beautify. I stayed on Presentation and left the model dropdown on Lite.

Prompt pasted almost as written: pitch deck for ToolifyNav, six to eight slides, cover line "Find the right AI tool faster," problem/solution/how-it-works/traction/ask, keep it concrete. Generate.

Credits went 400 → 349. Fifty-one credits for one seven-slide run. Against four hundred that is still several drafts, not unlimited evenings of regenerating everything.

What the seven cards actually said

Slide 1 opened with Find the right AI tool faster. under a darker header, subtitle ToolifyNav · Curated AI directory & hands-on evaluations.

Slide 2 originally titled itself AI tool discovery is noisy, vague, and expensive in time — three columns: VELOCITY, MESSAGING, EFFORT. Weekly launches, interchangeable marketing, costly trial runs. Close to the brief, a little wordier than I wanted on the heading.

Slide 3: ToolifyNav turns a crowded market into a shortlist you can trust — curated categories, hands-on reviews under my name, comparisons. Slide 4 walked browse → compare → read reviews → submit under Start broad, then narrow to the tool that fits.

Slide 5 is where the model got creative. Building a directory with free reach and paid visibility, plus an SVG bar chart of "Listing Share %": Standard Free 76%, Featured Category 16%, Premium Highlight 8%. Those are not my analytics. Illustrative filler. Fine for a draft stage; I would not paste that chart into a real investor mail.

Slide 6 put Free Listing vs Premium Highlight in a table. Slide 7 closed with Help people find the right AI tool faster. and toolifynav.com · Founded by Leo Liang. For a six-sentence prompt, that is a usable skeleton — not empty boxes with "Insert benefit here."

Edit, Present, then a free PDF

I rewrote the Slide 2 heading to Problem: AI tool spam and vague marketing — the line from my original brief. The three columns stayed put; only the title changed. It stuck when I clicked away. Opened Styles, flipped through a few presets, landed on something dark labeled Keynote Photography. Whole deck shifted in one shot. No upgrade wall for either change.

Present mode went fullscreen. Arrowed through the cards with the keyboard. The rewritten problem title was still on card two; the dark theme held. For a free account that is the whole loop I care about — generate, touch one thing, stand up and click through it.

Share → Export → PDF. Banner: Your PDF is being generated. It's safe to close the share panel. Download link in about eight seconds. Free. The pages carry a Made with Gamma badge. That is the free-tier tax. PPTX, Google Slides, and PNGs were listed in the same drawer — I did not open them in desktop apps, so I will not pretend about fidelity. Same for custom domains, the API, team seats, and the paid image models.

Four hundred credits and the upgrade ladder

Upgrade modal, Individual tab, annual toggle ("Save up to 28%"):

• Free $0 — 400 signup credits, up to 10 slides per prompt, exports with Gamma branding
• Plus $9/seat/mo annual ($108/yr) or $12 monthly — 1,000 credits/mo, up to 100 slides/prompt, remove branding
• Pro $18/seat/mo annual ($216/yr) or $25 monthly — 4,000 credits/mo, custom branding, analytics, up to 10 custom domains, API, premium image models
• Ultra $90/seat/mo annual ($1,080/yr) or $100 monthly — 20,000 credits/mo, advanced models, up to 100 custom domains

If you are a solo maker who needs a pitch or product walkthrough tonight and you refuse to open PowerPoint first, this free path is the rare one that actually finishes the loop: signup, generate, edit, present, PDF. If you need a clean export without their badge, or you regenerate long decks all week, the meter and the watermark push you toward Plus.

I did not burn the rest of the 349 credits trying every theme. One real deck was enough to see the shape of the product.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/gamma (official website, verified during review).

Getting there had been almost suspiciously easy. scisummary.com/register, email, password, confirm, terms checkbox. My review Hotmail, no Google. No captcha, and no verification email at all. I checked Inbox, Other and Junk out of habit and there was nothing to click. It dropped me straight into a dashboard with a modal that said "Explore SciSummary — Unlock Full Access with Your Free Trial". I closed it, uploaded a PDF, left the length slider around 400 words, and hit Summarize. Submitted at 16:11:40, summary on screen at 16:12:23. About 43 seconds.

Then the fade.

I write code for a living, so when a page hides text behind a gradient my first reflex is to check what the page actually delivered. The whole summary was there. That's how I graded it; a normal free user would only see the one sentence.

The paper I fed it

I didn't want to grade a summarizer on a paper I half understood, so I made a two-page fake agronomy paper where I know every number. Straw mulch versus bare soil, morning versus evening drip, field tomatoes, 36 plots in 12 blocks, 14 weeks, water held at 310 L a week. A table and a bar chart. Soil moisture 18.2% bare, 24.7% with mulch. Yield up 21%. Evening drip gave no significant yield gain (p = 0.41) but cut blossom-end rot from 9.8% to 6.1%. Hail took two harvests in block 7. Probes only at 15 cm. Don't extrapolate to plastic mulch or sandy soil.

The first version that showed up at 16:12 was a headed list: Objective, Methodology, Key Results, Limitations, Conclusion. The numbers were right. 18.2 to 24.7, the 21%, p = 0.41, 9.8 to 6.1. I was ready to be impressed, and then I read the Limitations. "Other factors like pest incidence or soil nutrient levels were not addressed." My paper never says that. It's filler, the kind of limitation you could paste under any field study. It also called blossom-end rot "disease control" (it's a calcium problem, not a disease), and it skipped the hail and the 15 cm probe depth, which were the actual limitations.

Then it rewrote itself

When I looked at the same summary slot again around 16:14, it held different text. A 403-word block of plain prose, no headings.

I don't know what runs behind the scenes. This version was good, though. Properly good. Every number was right, including the p = 0.38 for mulch alone on blossom-end rot, which is the kind of detail that tells you it read the results paragraph instead of the abstract. It said "evening irrigation produced no additional yield benefit beyond straw mulch", which is exactly the sentence I'd want a summarizer to get right, because the lazy reading of my abstract is "evening drip helps". It kept the hail in block 7. It kept the 15 cm probes. It even kept "about 11 person-hours per 1,000 m² were excluded" from the economics, and ended on "Results should not be generalized to plastic mulch or sandy soils."

So the engine is careful. The first draft wasn't, the second was, and I only know that because I happened to look twice.

One summary, then a wall

I tried one edit. + NEW SUMMARY, mode set to Key Points, submit.

"We could not complete your summary: You must activate your FREE trial or have an active subscription to generate summaries."

That was it. One summary per free account, and you can read about a sentence of it. What bugs me more than the limit is that nothing said so beforehand. No counter, no "1 of 1 used", nothing near the upload button. You find out by failing.

The rest of the app is behind the same door. Chat and Figures have lock icons and open the trial modal, so I never got to ask it about the table or the bar chart, which was half the reason I put them in. Podcast and Slides show up in the mode dropdown with "(Activate Trial to Use)". The two open tabs were a mixed bag. References got the title right, showed Identifier "NO_DATA", no authors and 0 references (fair, my fake paper has no reference list), with a note that the metadata "may not be complete". Recommendations sat on "Loading recommendations..." and never finished, probably because a made-up paper has nothing to match against.

I didn't get anywhere near the API or the MCP server, DOI or arXiv import, bulk runs, Zotero, or a long scanned PDF.

Four dollars a month, card first

Activate Trial sends you to Stripe Checkout: "Try SciSummary Premium", "7 days free", "Then $7.00 per month starting October 2, 2026". There's an annual toggle, "Save $36", which gets it to $4.00 a month. The trial asks for a payment method up front (card, US bank, Klarna, Cash App, Apple Pay, Link, Amazon Pay). I backed out without entering anything.

Where the plans live is its own small mess. scisummary.com/pricing is a 404. The plans are only on the homepage: Student at $0 for the first month with code STUDENT26, Pro at $4/mo billed $48 a year or $7 monthly, and the 7-day trial with 30,000 words, 5 AI figure analyses, 100 chat messages and 50 imports.

Four dollars is cheap for a summarizer that gets p-values and caveats right on the second pass, and if you're a grad student working through a stack of PDFs every week I think it's worth a trial, card and all. If you wanted a free tool, this isn't one. For free PDF answers you can actually read on screen, Humata's free tier was a very different afternoon. The listing is on SciSummary's ToolifyNav page if you want the rest of the details.

Final rating: 3 out of 5.

Sources: https://toolifynav.com/item/scisummary (official website, verified during review).

Signed up at app.humata.ai/signup with my Hotmail address. Email and password. First name Leo, last name Liang. No Google button, no Apple path. Verification mail from Humata.ai Support (noreply@humata.ai) hit Outlook Focused in about five seconds — not Junk this time, which already felt better than half the tools I review. Clicked the supabase.humata.ai link, landed on Free. Dashboard said $0. No card form. I could upload immediately.

Eighteen seconds, then the wrong click

I dropped in sample-paper.pdf — one page covering scaled dot-product attention, a toy copying-task score, and the softmax formula. The UI showed STEP-ENQUEUED, a progress bar around 70%, then SUCCESS with processing_engine_version: 3. Wall clock from drop to ready: about 18 seconds. Short enough that I did not go make tea.

The file table has a small trap. Click the row and you just toggle the checkbox. The thing that actually opens chat is a separate Ask link on the right. I missed it once and sat there wondering why nothing happened. Muscle memory from every other SaaS table failed me.

Worse: if you jump to /ask/file/:id while indexing is still running, you get a flat "Uh-oh. This file failed to load" instead of a waiting state. Wait out the pipeline and it opens fine. Still a sharp edge for anyone who bookmarks the URL too early or double-clicks Ask during ingest.

Once it loaded, the layout was chat on the left, PDF on the right. Humata had already written a short executive summary of the note and parked three suggested prompts under it. Composer modes along the bottom: Fast, Visual, Thinking. I stayed on Fast for the whole session. Did not poke the other two — that is a different afternoon.

Two questions, then a follow-up

First ask: "What exact-match accuracy did the Transformer reach on the copying task?"

About 2.5 seconds later: the Transformer reached 98.4% exact-match on the toy copying task, with a citation badge [1] tagged to page 1. That lines up with the PDF sentence about a 2-layer Transformer with 4 heads after 8k steps. No soft hedging. No invented second decimal.

Second ask: "What is the attention formula given in section 2?"

About 2.0 seconds. Answer came back as Attention(Q, K, V) = softmax( (Q Kᵀ) / √dₖ ) V, again with [1]. Clicking the badge snapped the right-hand viewer onto the formula. Yellow highlight covered the math and the temperature sentence — I counted roughly 18 highlight spans on the react-pdf canvas. That click-to-source loop is the whole product pitch, and on a clean digital PDF it actually worked.

Follow-up without restating the setup: "How does this compare with the LSTM under the same budget?" Another ~2.0 seconds. It kept the thread — Transformer 98.4% vs LSTM 71.2% under the same training budget. No amnesia between turns. For a free-tier chat over a one-pager, that is the bar I wanted.

I did not upload a second file. Did not try Excel or YouTube ingest even though the empty dashboard advertised both. Did not feed it a scanned photocopy. This run stayed on born-digital text where the parser already had characters to chew.

Free means sixty pages and a soft answer cap

Upgrade modal on Free: $0, up to 60 pages, up to 10 answers. Settings copy elsewhere talked about 100 questions. Same product, two ceilings written in different rooms. I did not burn the free quota to the wall, so I cannot swear which number wins when you actually hit the limit — only that both UIs exist and Free is clearly a sample lane, not a semester of literature review.

Paid ladder from the same modal: Student $1.99/mo (200 pages), Expert $9.99/mo (500 pages, 3 users), Team $49/user/mo (5,000 pages). OCR for scanned documents and images sits on Team. If your pile is phone photos of paper or flattened fax archives, Free will not save you. That price jump is the real gate for labs sitting on legacy scans.

Who this fits in practice: solo researchers and students with clean digital PDFs who want answers they can click back into the page. Who should look elsewhere: anyone living in scanned archives without a Team budget, or anyone who needs a full document management system with retention and legal hold. I did not test team sharing, SSO, enterprise SOC-2 claims, multi-hundred-page books, or mobile.

The core loop held. Signup without a card. Ingest in under twenty seconds. Sub-three-second answers that matched the paper. Citation highlight that was not decorative. The free caps, the checkbox-vs-Ask miss, and the early-open error are the tax you pay before Expert money.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/humata (official website, verified during review).

Signed in with my Hotmail address. No Google SSO. Clipdrop does magic links: “Check your mail inbox, we have sent you a link to sign in.” Inbox stayed empty. Junk had it — noreply@clipdrop.co, timestamped around 01:18 UTC on 21 Sep 2026. Clicked through, landed on Free. No card. No fake trial checkbox. That part I like. Deliverability into Junk is a small slap, but at least the link worked.

One file at a time, then downscale

I dropped test-product.jpg (1200×872 watch on gray) into Remove Background. Tried a small batch first out of habit — same file a few times, honestly, just to see the queue. Yellow banner: Free users can only process images one by one. Buttons: Remove 3 or Subscribe. I cleared the extras. Batch is clearly a paid story.

Next modal was worse for store assets: Image too large for the SD Mode. Side-by-side — SD 1024×744 vs HD 1200×872. Free path is Downscale & Continue. I took it. Already annoyed. A 1200-wide product shot is not exotic. Forcing SD before you even see the mask is an upsell with a polite smile.

Then Cloudflare Turnstile slid into the bottom-right of the editor: “Verify you are human.” Inside the canvas. Mid-job. Jarring. After I clicked through, inference ran. I get why freemium tools fight bots. I still hate solving a captcha between “upload” and “result.”

The cutout was the good part

About 3–4 seconds later the watch was on a checkerboard. Strap pinholes open. Buckle edges clean. No ugly halo on the bezel. I poked the eye toggle — before/after held up. Download gave me test-product.png, 184 KB, 1024×744, real alpha, no watermark. That core loop is why Clipdrop still earns a seat for “I need a cutout tonight.”

This is the part that almost pushed me higher. Watch hardware usually shreds cheap removers — mushy holes, melted clasps, white fringing on the crystal. Clipdrop’s mask did not. For a free-tier Remove Background pass, that quality is the product.

I did not run a second photo. Did not try hair, glass bottles, or people. One stubborn watch. Enough to trust the mask on that class of shot; not enough to crown it universal.

Relight: toast, then a React crash

I wanted the second pain fixed too — relight the cutout so the face looked less flat. Toolbar had a Relight shortcut next to the eraser. Clicked it.

Spinner. Another Turnstile. Blank canvas. Red toast: Error during the relight process, please try again. Cool.

Tried standalone /relight with a fresh upload. Dropzone looked disabled. Then the whole page fell into an unhandled React boundary: An error occurred. Sorry for the inconvenience. Go to home. Raw i18n keys leaking in the chrome — /header/item/relight, /footer/a/home, /sessionmanager/button/account. That is not soft-launch roughness. That is a broken tool page shipping translation keys to paying-attention users.

Poked /cleanup briefly. Upload handlers felt inconsistent — hidden inputs, flaky drop state. I stopped. Not running Upscaler, Text Remover, Uncrop, or Replace Background on this pass. Relight had already crashed twice; chaining from Remove Background was dead. The free pipeline was leaking enough.

Where did my 20 free runs go?

Pricing copy says free Background Removal is something like 20 / 24h. Account menu just said Current plan: Free and Subscribe. Settings: same Subscribe CTA. No number. No bar. No “17 left.” You fly blind until the wall, I guess. One of my three fail-able checks was “credit meter visible after one run.” That check failed hard.

So the score lands here: one excellent free remove-bg path (after downscale + Turnstile), watermark-free PNG you can actually download — then Relight dies, chaining dies, and you cannot see remaining credits. That is a 3, not a 4. Keep Clipdrop for the cutout. Do not plan a multi-tool lighting workflow on Free until Relight stops throwing React boundaries.

I still have the PNG on disk from this session. Opening it locally confirms a real alpha channel, not a fake white background. That download is the reason I did not abandon Clipdrop after the Relight crash — the cutout job finished and paid out a usable file.

Final rating: 3 out of 5.

Sources: https://toolifynav.com/item/clipdrop (official website, verified during review).

Quick scoreboard from the published ToolifyNav reviews:

• Grammarly — 4/5
• QuillBot — 4/5
• Wordtune — 3/5

Official sites: grammarly.com, quillbot.com, wordtune.com.

What I was trying to do

Three different jobs, on purpose, so the tools could fail in different ways:

• Grammarly: fix deliberate grammar mistakes in a 56-word sample inside the Free web Docs editor after email/password signup with Hotmail (no Google SSO).
• QuillBot: rewrite a 72-word work paragraph anonymously and see how far Free paraphrase modes go before Premium.
• Wordtune: turn a hybrid support/status-page request into usable line-level alternatives after email magic-link signup.

I did not start trials, enter cards, or invent results I did not see. Full session logs live in the three review posts linked above.

Grammarly Free — last-mile grammar, with one weird accept

On Grammarly’s Free web Docs editor, suggestions showed up a few seconds after paste. The panel showed Overall score 55, All 10 / Grammar 9 / Clarity 1, plus a Pro suggestions count of 1. Live flags I saw included Verb problem, Correct pronoun usage, Rephrase, Use the right word, and extra hits on and / sound / visitor / your / feel.

I applied two live suggestions: weather → whether, and a “Correct pronoun usage” accept on its that wrote `it` instead of the it's I expected. After those accepts the score moved to 61 and the text visibly contained “whether.” No card was required. Loud Pro CTAs stayed in the header (Try Pro for $0) and on the plan screen (Try Pro for 10 days free! declined via No, thanks).

Hotmail worked despite the “Work or school email” label. Outlook verification landed in about a minute after one mis-ordered code entry showed “Invalid code. Please try again.” Full-sentence rewrite, AI detector, and plagiarism stayed marked Not available on Free in the plan comparison I opened. Free plan copy I saw highlighted Brainstorm with AI chat and Write without mistakes.

Pick Grammarly when you want grammar/clarity flags inside a Docs-style editor and you are willing to read each accept before applying it.

QuillBot Free — anonymous paraphrase with a hard ceiling

QuillBot was the only one of the three that let me finish the core rewrite without an account. I dismissed the Google prompt on the homepage, opened Paraphraser via the product’s left rail (a direct /paraphraser guess showed “It appears you are lost…”), and ran a 72-word project-tracker paragraph.

Standard returned a 79-word colored diff effectively immediately — for example “tried three different project trackers” became “experienced with three alternative project trackers,” and the slipped-deadlines line shifted toward “updates remaining truthful.” Fluency refreshed without a payment prompt (72 words, 5/5 sentences) and leaned readable (“the board mirrored how we actually ship”). Formal hit the gate: Unlock Premium Modes, paraphrase button disabled.

The Premium comparison page showed Free at $0: paraphrase up to 125 words, 2 paraphrase modes, Humanize up to 125 words / 6 uses/day, plus basic grammar/summaries and limited AI Detector. The page also warned that limits may vary across platforms.

Pick QuillBot when you want a short anonymous rewrite and a visible Free vs Premium boundary.

Wordtune Free — fast suggestions, wrong voice on first apply

Wordtune redirected anonymous /editor traffic to signup. Email magic-link worked; no Google SSO, no card. Rewrite suggestions arrived in about 2–3 seconds with an easy-to-scan inline card. The first applied Rewrite turned my status-page request into instruction copy that started “Please provide…” instead of finished customer-facing status text. A more status-page-ish alternative existed on the same card (“My status page should state…”), but the first click was the bad one.

Formal also returned in ~2–3 seconds and stayed instruction-like (“Provide a clearer status page sentence that informs them…”). The Free meter dropped from 9 to 8 daily AI Generations left after those two actions. Visible Basic Free limits on the upgrade screen: 10 Rewrites & AI suggestions daily, 3 AI summarizations/month, unlimited spelling corrections and grammar checks. Advanced showed $6.99/mo billed yearly; Unlimited $9.99/mo billed yearly. Trial and Unlock All upsells were loud; I did not start a trial.

Pick Wordtune when you already have a sentence and will scan the suggestion rows carefully before applying the first rewrite.

How I would choose (same week, same hands)

• Need grammar cleanup inside an editor, no card: Grammarly Free — just do not blindly accept pronoun fixes after the its → it surprise.
• Need a short rewrite with zero signup: QuillBot Free — stay under 125 words; Formal is Premium.
• Need line-level alternatives and you already expect to babysit the first suggestion: Wordtune Free — meter and trial CTAs are part of the product surface.

If two tools are already on your shortlist, the pairwise pages built from the same evaluations are live:

• Grammarly vs QuillBot
• Grammarly vs Wordtune
• QuillBot vs Wordtune

What I did not test

Browser extensions, mobile/desktop apps, Google/Apple SSO checkout paths, paid Pro/Premium/Advanced behavior after a card, QuillBot Humanize and AI Detector limits in practice, Wordtune Casual/Shorten/Expand/Ask AI, and any hard exhaustion of Free meters beyond the counters I saw. Scores stay tied to the Free paths above — not to marketing pages I did not run.

Sources: https://toolifynav.com/item/grammarly-vs-quillbot-vs-wordtune (official website, verified during review).

Hotmail got in anyway

Signup path was plain: grammarly.com → Sign up → email/password only. Address: a plain Hotmail account. No Google. No Apple. No SSO. The field label said “Work or school email,” which usually means consumer domains get a polite rejection. Hotmail went through. No enterprise bounce, no “use your company domain” wall.

Verification code hit Outlook web in about a minute. All-folders search; message sat in Inbox, not Junk. Grammarly’s split OTP boxes bit me once — I mis-ordered the digits and got “Invalid code. Please try again.” Clicked the first box, retyped the full code clean. Instant verify. No Cloudflare puzzle. No card wall before the editor. That part was boring in a good way.

Onboarding asked what I was working on. I picked Other projects and Articles or blog posts. Right after that, the modal: “Try Pro for 10 days free!” with a fat “Try Pro for $0” button. I hit “No, thanks.” Did not start a trial. Did not enter a card. Free path stayed free.

Fifty-six words, score 55, ten flags

Inside app.grammarly.com I went Docs → New doc and pasted the whole sample in one go:

> I been working on a landing page copy all morning and its still sound vague. Their is three benefits we want visitor to notice, but the sentences keep dragging on and on without a clear call to action. If your reading this draft, can you tell me weather the tone feel too salesy or just confusing?

Word count blinked 0, then settled at 56. Suggestions showed up in a few seconds — not instant, not slow. Overall score landed at 55. Right sidebar: All 10 / Grammar 9 / Clarity 1, and a separate Pro suggestions count of 1.

What actually lit up against the traps I planted:

• I been — Verb problem
• its — Correct pronoun usage
• Their is — Rephrase
• weather — Use the right word
• plus and punctuation, sound subject-verb agreement, visitor wording, your word choice, feel subject-verb agreement

That is the core Free job I came for: paste messy copy, get a pile of concrete underlines, leave without paying. It did that. I did not need to invent a chat prompt to ask “please fix my grammar.”

The weird accept: its → it

I applied two cards only.

First: weather under “Use the right word.” One click → whether. Immediate swap on the canvas. The draft really contained “whether.” Homophone fix worked the way you expect.

Then its. Label: Correct pronoun usage. The line was ...and its still sound vague. In my head the fix is it's. The accept wrote `it`. Sentence became ...and it still sound vague. Possessive-versus-contraction mess traded for a bare pronoun that still leaves sound wrong. I kept the apply because that is what Free actually did on click — not what the label implied. Odd parsing for a product that has seen that typo forever.

After those two applies, the overall score moved 55 → 61. No lag. Surrounding formatting stayed put. I left Their is, I been, and the other flags alone so I could see the score move on a minimal edit pass.

Try Pro for $0, and what Free actually gets

The editor header kept “Try Pro for $0” pinned while I worked. Opening the plan screen brought the same pitch again: “Try Pro for 10 days free!” / “Try Pro for $0” / “No, thanks.” Loud. Consistent. Easy to decline if you look for the small link.

Comparison copy I saw on that screen:

• Free includes Brainstorm with AI chat and Write without mistakes
• Marked not available on Free: Rewrite full sentences in a click, Detect AI-generated text, Catch accidental plagiarism

No daily generation meter ticking down the way some rewrite tools do. Just the 56-word count and the writing-quality score. For basic grammar cards, that felt calmer than a burn-down counter — the Pro banner is still the noisy part.

I did not install the browser extension, open desktop or mobile apps, try Google/Apple SSO, start checkout, run the AI detector, try an AI rewrite/humanizer, run plagiarism, or fire full-sentence rewrites. This session stayed web Docs + Free suggestion cards + the plan screen.

If you need a dedicated canvas to catch verb mess and weather/whether without pasting into a chat, Free still does that without a card. Read every accept — the its → it swap is why I will not call this a 5. The trial CTAs will keep yelling. That combination is a 4 for me.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/grammarly (official website, verified during review).

Email signup, then a blank editor

Anonymous editor access was a dead end. https://app.wordtune.com/editor bounced straight to Wordtune Workspace signup. Fine. Continue with Email, Hotmail review address, Continue. Magic link landed in the already-open Outlook inbox. The first click on the visible link text did nothing useful, so I opened the exact URL from the email in the browser. That signed me in. No password, no Google SSO, no card, no trial, no Cloudflare. Landed in a fresh /editor/write/... document.

Rewrite in two to three seconds

Exact before text:

> Our support team gets the same webhook timeout ticket every week. The docs say to retry with exponential backoff, but customers still open chat after the third failure. I need a clearer status page sentence that tells them we already queued the event and they should wait fifteen minutes before escalating.

Rewrite. Loading skeleton. Suggestions in roughly 2–3 seconds. The editor selected only the trailing “I need…” sentence — not the whole paragraph. Good instinct for a line-level tool.

The card had a few usable rows. One of them looked more status-page-ish (“My status page should state…”). I clicked the first row anyway, because that is what most people do when they are in a hurry.

Applied result:

> Our support team gets the same webhook timeout ticket every week. The docs say to retry with exponential backoff, but customers still open chat after the third failure. Please provide a clearer status page sentence that tells them we have already queued the event and they should wait fifteen minutes before escalating.

“I need…” became “Please provide…”. Fluent. Also wrong for the job. It rephrased my request as an instruction to someone else instead of writing the status-page sentence I asked for. Fast alternatives, yes. Finished customer copy on first click, no. I would want a second pass before shipping that line.

Formal cleaned the grammar, not the intent

Formal. Another ~2–3 seconds with a loading skeleton. First Formal row applied:

> Our support team gets the same webhook timeout ticket every week. The docs say to retry with exponential backoff, but customers still open chat after the third failure. Provide a clearer status page sentence that informs them that the event has already been queued and they should wait fifteen minutes before escalating.

Cleaner. More formal. Still an instruction — “Provide a clearer status page sentence…” — not a banner customers can read. Tone worked mechanically. It did not fix the mismatch.

Proofread on the right also showed “clearer → clear” and later “chat → Chat”. I left those alone. I also skipped Casual, Shorten, Expand, Synonyms, Ask AI, Enrich, the extension, and anything mobile. This run stayed on Rewrite → first row → Formal → first row.

Nine left, then eight, then the upgrade modal

After the rewrite card: “9 daily AI Generations left” plus “Unlock for free.” After Formal: “8 daily AI Generations left.” The meter is not shy.

Upsell copy was everywhere I looked: “Experience Wordtune Premium — start your free 3-day trial now. Start trial,” “Unlock All,” “Start a Free Trial.” I did not start a trial. I clicked Unlock All to read the yearly view (Pay yearly (-50%) / Pay monthly):

• Basic $0 (Current Plan): 10 Rewrites & AI suggestions daily; 3 AI summarizations/month; unlimited spelling corrections and grammar checks.
• Advanced $6.99/mo billed yearly: 30 Rewrites & AI suggestions, 15 AI summarizations, plus unlimited AI recommendations, spelling/grammar, and Ask AI prompts.
• Unlimited $9.99/mo billed yearly: unlimited rewrites & AI suggestions, summarizations, spelling/grammar, vocabulary/clarity/fluency extras, Premium support, Ask AI prompts.

I didn't burn the Free meter to zero. The counter and CTAs still make the daily cap feel close early.

If you already know the sentence and want quick alternate phrasings inside an editor, Wordtune is usable on Free. Suggestions arrive fast, the inline card is easy to scan, and Formal actually changes the register. If you are hoping the first Rewrite click will invent finished status-page copy from a hybrid “here’s the problem / I need a sentence” paragraph, this session did not deliver that. Review the rows. The better status-page-ish option was visible; I just happened to click the instruction-like one first.

No publishing from Wordtune in this run. No paid-plan behavior beyond the modal. For line-level rewrites with a tight daily Free meter, this is a 3.

Final rating: 3 out of 5.

Sources: https://toolifynav.com/item/wordtune (official website, verified during review).

Seven seconds on fasting and HbA1c

Exact prompt:

“What did 2023–2024 systematic reviews and randomized trials find about intermittent fasting and HbA1c in adults with type 2 diabetes? Please cite studies, report HbA1c changes and limitations, and distinguish direct trial evidence from broader reviews.”

UI: Researched 7s. First useful prose landed almost immediately while the rest kept streaming. Not diet-blog mush. It put numbers on the page: a 2023 RCT meta-analysis with SMD −0.08 (95% CI −0.20 to 0.04; P=.19); INTERFAST-2 intermittent fasting −7.3 mmol/mol (~−0.6 percentage points) vs control +0.1; EARLY 5:2 meal-replacement −1.9 points at 16 weeks vs metformin −1.6 and empagliflozin −1.5. It called out limitations and bundled components in the same breath.

Source drawer: 24 sources. Links tab showed PMC, PubMed, Diabetes Care and friends. That is the product working — I stayed on one screen instead of opening ten tabs myself.

A card also sat over the answer: “This question looks difficult” with Upgrade to Pro. The answer still finished on Free/Incognito. Soft upsell, not a hard stop. Mildly annoying. I kept reading.

The follow-up that became a table

I asked it to rebuild the evidence:

“Turn the evidence into a concise table comparing the 2023 meta-analysis, INTERFAST-2, EARLY, and the 2024 16:8/14:10 trial: sample, duration, HbA1c change, and what each actually proves. Flag any study where the intervention bundled meal replacements or medication changes.”

Researched 8s. Sources climbed to 30. The table held the caveats: insulin-dose adjustment in INTERFAST-2; meal replacements, calorie guidance, contact, and education bundled into EARLY. That was the moment I stopped testing “can it write” and started trusting it for structure. Prose-to-table without dropping the footnotes is rare.

I did not independently recalculate those medical effect sizes. I checked that the citations looked concrete and that pooled evidence was separated from direct trials. Good enough for this session; not a systematic review of my own.

Thirty-six seconds, then Chinese

Second query, different domain:

“What evidence from 2023–2025 studies supports or challenges the claim that remote work changes software-engineer productivity? Summarize at least three studies with methods and measured outcomes, and distinguish causal from correlational evidence.”

Researched 36s. Over four times slower than the diabetes ask. Fine. Waiting.

Then the whole answer rendered in Chinese even though the prompt was English. The synthesis underneath still looked serious across 25 sources: Gibbs/Mengel/Siemroth (2023) full-remote IT setting with output/hour down about 8–19%; Bloom/Han/Liang (2024 Nature) hybrid RCT, 2 days/week at home, no significant performance/code-line damage and lower attrition; BLS industry stuff labeled correlational. Useful distinctions. Wrong language. That glitch is the kind of thing that makes you reread the thread twice before you trust the next one.

I clicked the Nature citation anyway. It opened “Hybrid working from home improves retention without damaging performance” (12 June 2024). Nature’s own cookie dialog. Real link. So the retrieval path worked even when the language router tripped.

Another Upgrade to Pro card on this difficult question, plus a “Pro Free preview of advanced search enabled” banner during the follow-up / second query stretch. Still no hard wall.

Pricing as shown, and who this is for

Personal pricing modal, annual billing:

• Pro — $17/month. Expanded Computer access, 4,000 bonus credits, deep research, top models, model selection, polished documents/apps, higher limits/memory.
• Max — $167/month. 35,000 bonus credits and 10,000 monthly credits.
• Free — referenced as the base (“Everything in Free”). Account shell said Free plan. Modal did not show a separate numeric Free allowance, so I never watched a daily counter tick down.

No purchase flow. I also skipped file upload, Collections, exhausting a daily quota, and email signup/OTP.

If you need fast cited synthesis and a table rebuild on Free without creating an account first, this session delivered. If you need predictable language on every hard prompt, or a clear Free allowance before you commit to a research day, those are still open questions from what I saw.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/perplexity (official website, verified during review).

Email signup worked. Verification landed in Outlook Junk from Aman@explainpaper.com. After I confirmed, the workspace opened. No Google SSO in this test.

Getting into the sample

The first click on the dashboard sample-paper row did nothing. Annoying. I opened the sample route directly: /read/u/sample. That worked. Built-in paper: Attention Is All You Need, page 1 / 15. I did not upload my own PDF.

The highlight that took a couple tries

I dragged this abstract span:

“mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions”

The first couple of native drags did not stick. No selection, no explain affordance. Then the rendered text layer finally caught it and the blue highlight showed up. Small thing. Still the kind of friction that makes you wonder if you are fighting the PDF canvas or your own mouse.

Understanding level sat on Undergrad. I hit Explain.

About five seconds later the side panel returned:

“The authors introduce a new type of neural network architecture called the Transformer that primarily relies on attention mechanisms to process information. Unlike traditional models that use complex structures like recurrent neural networks (RNNs) or convolutions, the Transformer simplifies the process by focusing solely on how different parts of the input data relate to each other through attention. This approach enhances performance and efficiency, particularly in tasks like machine translation, making it faster and easier to train compared to previous models.”

That is a real paraphrase, not a synonym swap. It named the Transformer, contrasted RNNs and convolutions, and kept the answer glued to the selected abstract instead of dumping me into a disconnected thread. No numeric speed claim. No citation. Fine for “what did this sentence mean.”

I caught myself relaxing. Five seconds is not instant, but it is fast enough that I kept reading the abstract instead of opening another tab. That is the product working.

Follow-up without leaving the margin

Under the card: Ask a follow-up question. I typed:

“Can you explain what “dispensing with recurrence and convolutions” means in practical terms for training?”

Another ~5 seconds. Different answer, which is what I wanted:

“Sure! "Dispensing with recurrence and convolutions" means that the Transformer model doesn't use recurrent neural networks (RNNs) or convolutional neural networks (CNNs), which are common in traditional models. In practical terms, this allows the Transformer to process all words in a sentence simultaneously rather than one at a time (as RNNs do). This parallel processing speeds up training significantly. Additionally, it avoids issues like vanishing gradients often encountered with RNNs over long sequences, making it easier for the model to learn from longer contexts without losing information.”

It moved from architecture summary into training implications — simultaneous tokens, vanishing gradients. Still said “significantly” without a number. I stopped there instead of hunting a second independent highlight.

Free vs Pro, as shown

Below the explanation: Upgrade to Pro for GPT-5. Math Explain was disabled and pushed an upgrade prompt. The highlight explain and the follow-up both ran on Free. No credit wall. No numeric free-credit counter anywhere I could see, so I never watched a balance tick down.

Pricing page (I flipped Monthly and Yearly):

• Free — $0/month, “Free for everyone, forever.” Unlimited highlight explanations, follow-up questions, Zotero import, basic AI models.
• Pro — $16/month monthly, or $12/month on yearly with a visible Save 25% label. 7-day free trial button. Math Explain, advanced models, whole-paper summaries.
• Teams — Contact us.

I did not start the trial or enter a card. On Free I also never saw a remaining-credit meter. This session never hit a wall. Day-20 of a lit review under Free is still an open question for me.

Who this is for, and where I stopped

If you are an undergrad, a developer reading ML papers, or anyone who keeps alt-tabbing jargon into ChatGPT, the Free highlight loop is the product. Stay on the page. Ask one follow-up. Move on.

I did not upload a local PDF, paste a PDF URL, import Zotero, run Math Explain, summarize the whole paper, or burn through any hidden Free cap. Selection friction and the dead first dashboard click are what kept this from feeling like a 5. The core explain path still earned the session.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/explainpaper (official website, verified during review).

Homepage claim: “100% Automatically and Free.” Also a banner I almost scrolled past: the standalone site moves to Canva and shuts on 1 December 2026, 9:00am CET. That colored the whole session. Hard to get excited about stacking free-tier habits on a tool with a hard sunset.

Dropping the photo in

No login wall. Upload page just worked. I picked my local JPG from the box and waited.

Roughly 2–5 seconds later the cutout was there. Not stopwatch-timed, but nowhere near my private “under 30 seconds or I bounce” bar. For a first anonymous pass that felt almost rude how quick it was.

The checkerboard preview kept the messy bits: speaker + cable, charger, clip, brush, pen, phone, card, watch, tube, glasses, little accessories. Thin stuff usually dies first. Here the cable, pen, and glasses frames stayed. No obvious leftover black rectangle where the background should have been punched out.

Still irritated me: a soft gray/white shadow/gradient along the bottom edge. Dark-on-black makes edges harder to judge, and that leftover would bother me on a real listing. I stared at it longer than the cutout itself. I did not try hair, glass, metal reflections, people, or a second photo — this was one stubborn flat-lay, and I left it there.

The download wall

Download menu was blunt:

• Preview 612 × 408 — Free
• Max 1200 × 800 — Unlock

No watermark on the preview. No card popup when I hit Free. I clicked the free Preview option twice. Menu stayed open. Nothing new showed up in my Downloads folder. Maybe browser weirdness on my side — I’m not calling it a Remove.bg outage. What I can say from the UI: free is explicitly 612×408, and 1200×800 is Unlock.

That resolution is a draft thumbnail. Not a zoomable store asset. I almost bounced right there. For a Shopify-style product tile you want sharper than this without paying. The cutout quality had me leaning toward a 4; the free export size yanked it back hard.

Color worked. Magic didn’t.

I poked the editor mostly to find the first paywall.

Background → Color: clicked the red swatch, checkerboard became solid red, objects stayed. No signup. Fine for a quick comp. Honestly the part of the free path I trusted most in this session.

Background → Magic: “Sign up to try.” So the AI background they advertise next to Color is gated the moment you poke it. I stopped. Didn’t make an account for this pass. Didn’t open Effects / Adjust / Design beyond that Color vs Magic glance.

Pricing (and the clock)

On remove.bg/pricing I saw pay-as-you-go at 3 credits for $3, subscription at 40 credits for $9/month, plus free copy about “1 trial credit to spend on AI edits” and “Basics, at no cost.” Same Dec 1 2026 expiry language as the homepage banner. Credits dying with the site is a real planning problem if you’re stocking up.

I didn’t buy Unlock. Didn’t run a card. Didn’t touch API, batch, or plugins. The pricing page mostly confirmed what the download menu already told me: free is a tease, Unlock is the product, and the calendar is not on your side.

Who this is actually for

If you want a five-second anonymous “will this flat-lay even separate?” check, or a free solid-color mock, Remove.bg still feels good — and the thin-edge pass surprised me more than I expected going in.

If you need free production resolution for ecommerce, or you’re planning credits past late 2026 without moving to Canva, this session was a no. Also listed on ToolifyNav if you want the directory card next to this write-up.

Fast cutout. Fiddly edges held up. Free export is cramped. The site is on borrowed time. I’m fine leaving it at three and moving on.

Final rating: 3 out of 5.

Sources: https://toolifynav.com/item/remove-bg (official website, verified during review).

I opened chatcsv.co on September 9, 2026 (UTC+8) to do a boring job: log in with email, upload a five-row CSV, ask one concrete question, then look at limits. I did not finish that job. That failure is the review.

Getting in

The homepage still sells spreadsheet-chat in plain language. Paste a file, ask questions, skip standing up a BI project. That pitch is clear.

It also says ChatCSV’s successor is Obvious, with a link out. The product feels sunset while the marketing copy still invites you in.

I used email login only. No Google. No GitHub. The UI said it sent a magic link that expires in 24 hours.

I checked Outlook Inbox. Empty of ChatCSV. I checked Junk Email. Same. A mailbox search for “chatcsv” returned no results.

I tried /login/email-password as a backup path. It was a 404. There was no password field on the login screen. Magic link was the only door I could see, and the mail never showed up.

Login status: not completed. Zero authenticated workspace. No assumption that a later email arrived after I stopped looking.

The demo upload

Before opening the site I wrote a tiny CSV at /workspace/review-chatcsv/sample.csv: five rows, columns id, date, amount, note. Coffee, lunch, snack, software, transit. Small on purpose.

The public /demo workspace opened with a Titanic dataset already loaded and a “what’s the data about” answer already on screen. That part is easy to understand. You land inside a working-looking chat without signing in.

I attached the sample CSV from the paperclip control. The UI showed a red error toast and reset to “0 files selected.” No file stuck. No progress bar that finished. No new dataset label with my filename.

I am not claiming anything about how ChatCSV handles that CSV. The upload failed in front of me. Metrics for this step: attempted uploads 1, successful uploads 0.

Queries that did not land

The preloaded Titanic write-up named columns such as PassengerId, Survived, Pclass, Name, Sex, Age, SibSp, and Parch. Fine as a canned explanation of a canned table.

I sent: “How many passengers are in each passenger class?” The UI briefly showed “Stop generating,” then returned to the prior view. No new user message bubble I could point at. No new result card I could measure.

I sent a second query: “Which passenger class had the highest survival rate?” Same behavior. Generating flashed. Nothing new stayed.

Latency: not timed. There was nothing new to time. Authenticated CSV query success: 0. Public demo new-answer success in this session: 0 of 2 attempts.

I do not treat the old Titanic blurb as proof that a fresh question works. It was already there when I arrived.

Pricing observed

The homepage FAQ says ChatCSV is free for everyone, and that after the Flatfile acquisition they went free. I did not see a numerical usage cap on the pages I opened.

I did not reach a signed-in billing page. I did not enter a card. I did not click a paid upgrade. Pricing claims here are FAQ copy only.

What kept it from a 5/5

The product may still work for someone whose magic link arrives and whose upload sticks. That is not what I got on this pass.

• Magic link: 0 messages in Inbox or Junk during the check window.
• Own-CSV upload on the public demo: failed (red toast, 0 files selected).
• Two new demo queries: no visible new answer after “Stop generating.”
• Positioning: the homepage treats ChatCSV as superseded by Obvious.

A free tool you cannot enter, and a demo that will not take your file, is not a 4/5 day. I scored the path I walked.

Not tested

Authenticated workspace. Password login (route 404; no field shown). Successful upload of sample.csv. Answer correctness on my own data. Charts, export, sharing, team features, API, deletion/privacy controls, billing. Mobile. I do not assume the Titanic demo equals a working personal workflow. I do not assume Obvious replaces every ChatCSV feature.

Final rating

Readable demo shell. Clear “it’s free” FAQ. And I could not log in, could not upload my file, and could not get a new answer to land.

If ChatCSV is winding down into Obvious, say that louder on the login path. If it is still meant to work, the magic-link delivery and demo upload need to work on a cold try.

Final rating: 2 out of 5.

Sources: https://toolifynav.com/item/chatcsv (official website, verified during review).

I opened Mem on September 9, 2026 (UTC+8). The marketing site redirected to get.mem.ai, while the core web application ran at mem.ai. I wanted the loop they sell: write a note, ask the Agent about it, edit it, then look at the free-plan numbers. Not a medical-evidence review.

Getting in

Email and password got me in. Then a verify link.

• I registered with an email address and password.
• Verification was required before entering the workspace. Mem sent an email to my Outlook inbox with a “Verify email address” link, which I opened to confirm the account.
• I did not use Google SSO or Apple SSO during authentication.

Once verified, Mem loaded a multi-step onboarding sequence. The interface prompted for:

• Referral source selection.
• An introduction to the Mem Agent.
• A prompt to connect Google or Outlook calendars.
• An upsell modal offering a 7-day Mem Pro trial, which stated “7 days free” followed by “$29/month.”
• A suggested first message to the Agent.

I skipped both the calendar integrations and the 7-day Pro trial. No payment method or credit card details were entered during the setup. The onboarding sequence adds several modal steps before you reach your first blank page, but once dismissed, the main workspace is uncluttered.

Capture and edit

Mem handles note creation without demanding immediate folder structures or metadata tags.

I created one private note titled “Intermittent fasting — research notes” containing:

• A defined research scope comparing time-restricted eating against standard calorie restriction.
• Specific questions about schedule variations, baseline health factors, and safety contraindications.
• An evidence checklist covering study population, comparator, duration, adherence, measurable outcomes, and source links.
• An explicit draft and review status label.

Mem automatically derived the note title from the first line of text. The note immediately populated in the left sidebar list, displaying “Private” and “Edited today by Account.”

For the edit test, I updated the note after running my AI query. I appended the sentence: Edited during review: added a source-verification reminder. The edit persisted immediately and remained stable after the page settled.

Ask your Mem

To evaluate search and retrieval, I used the main “Search or ask your Mem” bar.

I submitted the prompt: “What are the questions to verify in my intermittent fasting note?”

Mem Agent queried the active note in the workspace:

• It identified the specific document accurately.
• It directly surfaced the safety and contraindication questions from the body text.
• It summarized the scope and the evidence checklist clearly.

It summarized my own note. No citations, no source chips. Fine as memory. Not a research engine.

Pricing observed

I inspected the account tier limits and pricing options directly.

When I navigated to https://mem.ai/pricing from within the logged-in web app, the application returned an authenticated “Not found” error screen. However, visiting the marketing pricing page at https://get.mem.ai/pricing worked properly and displayed all four plan cards.

Here are the concrete plan tiers and limits published on the marketing page:

• Free ($0/month):

- 25 Messages to Mem - 25 New Notes per month - Task and project tracking for up to 25 notes - 15 open tasks/projects - 3 hours of Voice Mode

• Mem Plus ($9/month):

- 50 Messages to Mem - 50 New Notes per month - Task and project tracking for up to 50 notes - 25 open tasks/projects - 10 hours of Voice Mode

• Mem Pro ($29/month):

- 100 Messages to Mem - 100 New Notes per month - Task and project tracking for up to 100 notes - 50 open tasks/projects - 1 custom routine

• Mem Pro 2× ($49/month):

- 200 Messages to Mem - 200 New Notes per month - Task and project tracking for up to 200 notes - 100 open tasks/projects

During this test session, my total usage metrics were:

• Notes created: 1
• AI/Agent questions sent: 1
• Notes edited: 1
• Calendar connections: 0 (skipped)
• Pro trials started: 0 (skipped)
• Payment methods entered: 0
• Visible pricing cards: 4

The visible free allowance of 25 messages and 25 new notes per month gives users a concrete sandbox, while paid upgrades at $9, $29, and $49 scale volume predictably.

What kept it from a 5/5

1. In-app pricing route mismatch: Navigating to https://mem.ai/pricing returned an authenticated Not Found view inside the app. Pricing was only accessible by navigating to https://get.mem.ai/pricing. 2. Onboarding modal sequence: The setup flow presents several upsell and integration modals before allowing direct access to the note editor. 3. No citation interface in Agent recall: The Agent summarized note content accurately, but did not provide granular source citations or attribution tags in its output.

Not tested

To maintain clear test boundaries, the following areas were not tested during this evaluation:

• Google SSO and Apple SSO authentication options.
• Calendar synchronization with Google Calendar or Outlook Calendar.
• Email connectors, external integrations, bulk file imports, and data export features.
• Mobile applications (iOS and Android) and desktop push-to-remember workflows.
• Voice Mode audio recording and voice transcription.
• Multi-note task and project tracking pipelines.
• Custom automated routines on paid plans.
• Team workspaces, sharing permissions, and multi-user collaboration.
• Developer API endpoints, offline note editing, and account recovery procedures.
• Subscription renewal, payment processing, cancellation terms, and paid-feature gating post-trial.
• Importing external source documents or validating third-party medical claims.

Final rating

Capture was easy. The Agent found the questions I wrote. Limits on the marketing page were unusually concrete.

What kept it from a 5/5 were the broken in-app pricing URL, the modal gauntlet, and Agent answers with no citations.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/mem (official website, verified during review).

Most AI search tools answer science questions by summarizing whatever ranks on the open web. I wanted something narrower. Take one outcome-specific question, land on peer-reviewed studies, and see whether the product gives me an evidence trail I can actually inspect.

The query I ran on 9 September 2026 (about 09:06 Asia/Shanghai): “Does intermittent fasting improve insulin sensitivity in adults?”

Getting in: the email wall

Homepage copy was clean: “Research starts here,” “Consensus is the AI-powered academic search engine,” “Search & analyze 220M+ peer reviewed research papers,” plus “Transparent, reliable, and built to save you time.” Quick chips for “Find promising research questions,” “Search with filters,” and “Find the Consensus.” Search placeholder: “Ask the research...” Controls: Corpus, Deep +, Filter. No cookie banner.

I typed the question and hit submit, expecting at least a peek at results. Instead I got an auth screen: “Enter your email,” “Continue with Google,” “Continue with Apple.” No anonymous preview.

I skipped Google and Apple, entered no password and no card, and used the review Hotmail. Outlook showed a Consensus message with a 6-digit verification subject. After the code, onboarding asked “Which best describes you?” (I picked Other), “Bring your research with you!” (I’ll do this later), then “Ask research questions. Get cited answers.” (Ask my own). Sidebar allowance on entry: “5 / 5 Pro messages left.”

The search: 13 seconds to a Consensus Meter

I ran the same query again in the Corpus / Pro flow.

• About 5 seconds in: loading / scanning.
• About 13 seconds in: synthesis, Consensus Meter, and paper cards filled in.

Top of results showed “Results 52.3K.” The Pro step also said “Read Abstracts and PDFs 20.” I treat 52.3K as the displayed pool, not papers I personally read.

AI synthesis, verbatim: “Yes, intermittent fasting improves insulin sensitivity in adults, though effects vary by regimen, population, and duration.”

Consensus Meter N = 20: Yes 15 (75%), Possibly 3 (15%), Mixed 0, and a red segment of 2. The red legend text sat below the fold, so I am not naming that label beyond the visible 2/20 count.

Cards showed titles, years, authors, citation counts, journals, supporting-quote counts, methodology / journal badges, and PDF buttons, with a References pane. First cards included the Yuan et al. systematic review / meta-analysis on intermittent fasting and glucose / lipid metabolism, and a GRADE-assessed systematic review on insulin resistance, lipid profile, and inflammation in metabolic syndrome.

That combination — meter plus inspectable cards — is what made the question feel testable instead of chatty.

Paper detail and Study Snapshot

I opened the first paper. Overview: 24 March 2022; International Journal of Endocrinology; Q3 SJR; 101 citations; 3 influential; DOI 10.1155/2022/6999907. The abstract described a systematic review of clinical trials using random-effects or fixed-effect models. Visible outcomes included fasting blood glucose −0.15 mmol/L, glycated hemoglobin −0.08, insulin plasma −13.25 uIU, HOMA-IR −0.31, BMI −0.8 kg/m², body weight −1.07 kg, and waist circumference −2.08 cm, with confidence intervals in the UI. Tabs included Attachment, Evidence (1), Metadata, plus link/copy and download icons. I did not open the PDF.

Generating Study Snapshot showed “10 Study Snapshots left” and “You have 10 Study Snapshots left — you will get 10 more on October 8th. Upgrade to Pro for unlimited Snapshots, Pro Analysis, and more.” Fields returned:

• Population: Patients with metabolic syndrome
• Methods: Meta analysis
• Outcomes: Glucose metabolism, lipid metabolism, insulin resistance, body weight measures
• Results: Intermittent fasting reduced blood glucose, insulin, BMI, cholesterol, and improved insulin resistance.
• Badge: “Extracted 4/7 study attributes”

Fast and concrete. Also visibly partial.

Filter edit and allowance counters

I opened Filter, chose “Past 5 yrs” under Publish year. Range became 2021 — 2026, UI showed “Filter (2)” and “Apply 2 filters.” Open access and Exclude Preprints stayed untouched. After apply, the headline still read “Results 52.3K.” The filtered total never became obvious.

After the one Pro search, sidebar moved from “5 / 5 Pro messages left” to “4 / 5 Pro messages left.”

Pricing observed

Individual / Annual pricing page:

• Free — $0/mo. “Keep using basic paper search for free, without any AI analysis.” Basic Paper search; 15 Pro messages per month; up to 3 Deep reviews per month.
• Pro — $12/mo ($144 annual, “SAVING $96/YR”). Unlimited Pro messages; 15 Deep reviews per month.
• Deep — $45/mo ($540 annual, “SAVING $240/YR”). Unlimited Pro messages; 200 Deep reviews per month.

I did not pay. I also did not invent an explanation for this mismatch: pricing copy says 15 Pro messages/month on Free, while my new account sidebar started at 5 / 5 and dropped to 4 / 5 after one search.

What worked / what kept it from a 5/5

What worked: a 13-second Pro synthesis with a 20-paper meter; paper cards with DOIs and effect sizes in the UI; Study Snapshot structured in seconds.

What kept it from a 5/5 were the hard email wall before any results, the filter that left “Results 52.3K” unchanged, the snapshot stopping at 4/7 attributes, and the Free-plan 15 vs sidebar 5 Pro-message mismatch.

Not tested

Not tested: anonymous results (blocked by signup), Google / Apple / password login, Deep review, Ask Paper, PDF download/open, Citation Graph, API / MCP, Team / Enterprise, monthly billing toggle, payment, Open access filter, Exclude Preprints.

No assumption: 52.3K is not “papers analyzed”; the red meter label beyond the visible 2/20 segment; that the year filter changed the 52.3K headline; that 4/7 snapshot fields are complete.

Who it’s for

Students, clinicians, and researchers who already have a specific scientific question and want a rapid cited consensus check with abstract-level evidence, not a generic chatbot summary.

Verdict

Consensus turned my intermittent-fasting question into a measurable trail — meter, 20 abstracts, DOI-backed cards, and a snapshot — in one sitting on the free tier. Signup before search, sticky headline counts, partial snapshots, and the Pro-message allowance mismatch are why it stays a 4 for me. The full Consensus review lives with the ToolifyNav listing.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/consensus (official website, verified during review).

Pick your path first

Free Launch is $0 and requires the official Listed badge plus a backlink on your own site. It joins the standard weekly queue, which currently runs 14+ days.

Premium Launch is $9.90 one-time. No badge is required, and the listing publishes after payment with 7-day Featured placement labeled sponsored.

Highlight is $20 one-time for one week: a labeled highlight card in the directory feed plus homepage New Arrivals exposure.

Popular rankings stay organic. We do not sell Popular slots.

Path A — Free Launch (badge + Verify)

A free listing is published from the Review & Publish page, and only once the live badge check passes.

• Open toolifynav.com/submit and choose Free Launch.
• Paste your product URL and run Smart fill to prefill the listing, then correct anything the autofill got wrong. Editing the URL afterwards discards the autofill data.
• Complete the required fields: product name, Category, Sub-Category, 1–3 Tags, short description and a cover image. The GitHub repo and the long-form markdown body are optional.
• Copy the official Listed badge HTML from toolifynav.com/badge and place it on your site — the footer or an about page works.
• Submit the form. Free submissions are reviewed from the standard queue, so allow the 14+ day window before expecting a live page.
• Open the Review & Publish page for your submission and click Verify badge. A passing live check publishes the listing immediately; a failing check keeps it unpublished, so fix the badge snippet and verify again.
• Once published, your tool has a permanent /item/<slug> page.

Skip Free if you are not willing to put a badge on your marketing site. Use Premium instead.

Path B — Premium Launch ($9.90)

• Open toolifynav.com/pricing, or pick Premium Launch on the submit form.
• Pay once through checkout. It is a one-time payment, not a subscription.
• The listing publishes after payment clears. No badge is required on your site.
• You get a Featured badge labeled sponsored and Featured placement for 7 days.
• Your listing can link to your official site; that link is marked sponsored.

Path C — Highlight ($20 / week)

• Same auto-publish behavior as Premium, with a stronger labeled highlight in the directory feed for one week.
• Includes homepage New Arrivals exposure and a newsletter mention.
• Still one-time. Nothing renews on its own.

Badge requirements for Free

Free listings need a verified badge before they go live. Premium and Highlight publish automatically after payment and do not require a badge. Use the official HTML from the badge page, keep the hosted SVG URL and the native size, and do not restyle or recolor it.

What we do not offer

• Guest posts or paid link inserts into blog articles.
• Dofollow links. The badge link stays nofollow.
• Buying Popular ranking.

A hands-on review later is an editorial decision, separate from any launch plan. Editorial questions go to info@toolifynav.com.

Checklist before you click Submit

• The product is an AI tool people can actually open, not a vapor landing page.
• The cover image shows the product, not a logo stub.
• The category and sub-category match what the tool does.
• Free path: the badge is live and crawlable on your site before you verify.
• Paid path: you are comfortable with the sponsored label on the listing.

Start here

• Submit your tool: toolifynav.com/submit
• Compare plans: toolifynav.com/pricing
• Get the badge: toolifynav.com/badge
• Contact: info@toolifynav.com

Sources: https://toolifynav.com/item/how-to-submit-to-toolifynav (official website, verified during review).

My goal was narrow and practical: paste a standard work paragraph, get a usable rewrite without paying, toggle between the modes that are actually free, and see exactly where the free tier hits a wall.

What I actually did

I opened https://quillbot.com directly in Chrome. There was no cookie banner in the way. A Google account sign-in prompt popped up over the interface, but I dismissed it immediately and stayed logged out. The homepage presented the Paraphraser tool in the left rail alongside a prominent green “Upgrade to Premium” button.

Before running any text, I checked the official comparison page at https://quillbot.com/premium. The pricing grid showed a $0 USD/month Free column positioned next to Premium and Team Plan. The Free tier listed the following specific inclusions:

• “Paraphrase up to 125 words”
• “Paraphrase in 2 modes”
• “Fix basic grammar errors”
• “Humanize up to 125 words, 6 uses/day”
• “Generate basic summaries”
• “Access AI Detector (limited)”
• A footnote warning: “Limits may vary across platforms”

Next, I tried navigating directly to /paraphraser by guessing the URL. That threw a page error: “It appears you are lost…”. I backed out and clicked the left-rail Paraphraser link instead, which routed properly to /paraphrasing-tool.

Once the editor loaded, I picked Standard mode and pasted a neutral five-sentence paragraph about project tracking. QuillBot’s interface word counter clocked the input at 72 words. I hit the green “Paraphrase” button.

The output appeared in the right-hand panel effectively immediately. I did not run a stopwatch, so I will not claim an exact millisecond latency, but there was no noticeable lag. The right panel showed an output counter of 79 words across the five sentences, with colored diff highlights showing what shifted. The rewrite changed “tried three different project trackers” to “experienced with three alternative project trackers,” modified a sentence about calendar sync and duplicate events, and reworked “status updates stayed honest when deadlines slipped” into phrasing about updates remaining truthful when deadlines were missed.

To test the second free option, I switched the mode selector to Fluency. The panel refreshed on the spot without asking for a payment or a login. The right panel updated its counter to 72 words across 5 / 5 sentences. The Fluency pass focused on readability—for instance, changing “the board finally matched how we actually ship” to “the board mirrored how we actually ship.”

Then I clicked Formal to test a locked feature. The editor immediately switched to an informational panel explaining Formal mode with listed use cases (research papers/reports, essays, professional presentations, work/professional emails, cover letters, and technical/white papers). The main action button turned into “Unlock Premium Modes”, and the paraphrase trigger was disabled. That was the hard boundary of the free tier.

Finally, I checked the account onboarding flow. I opened the profile menu and brought up the login/signup modal. It offered single sign-on buttons (Continue with Google, Continue with Facebook, Continue with Apple) and a standard email signup/login path. The email form required three fields: Work or school email, Password, and Name. I stopped there without filling in details or submitting. No credit card was requested because I did not proceed past the credential form.

What worked and what annoyed me

The anonymous core flow worked cleanly. I did not need to sign up, hand over an email, or touch billing to rephrase a short paragraph. Standard mode was assertive, slightly expanding the text from 72 to 79 words to vary the vocabulary. Fluency mode was more conservative, holding the length at exactly 72 words while smoothing out the phrasing. The colored inline diff made comparing the changes fast.

What kept it from a 5/5 were a few clear constraints and minor interface hiccups:

• The 125-word limit is adequate for a standalone paragraph or an email snippet, but it will immediately block anyone trying to clean up a multi-paragraph article section or report in one paste.
• The direct /paraphraser route was broken in this session, showing the “It appears you are lost…” dead end instead of redirecting to /paraphrasing-tool.
• The Google SSO overlay appears immediately on landing, requiring an extra click to dismiss before getting to work.
• Only two modes (Standard and Fluency) are available for free; higher-value options like Formal are hard-gated behind the “Unlock Premium Modes” prompt.
• The UI exposes a word counter only; there is no character count displayed, so text length is strictly governed by word metrics.

Limits and Premium gates

Pricing listed Free at $0 USD/month with paraphrase up to 125 words in 2 modes, plus Humanize up to 125 words / 6 uses per day, basic grammar, basic summaries, and limited AI Detector. Formal mode never ran on free — it showed Unlock Premium Modes and disabled paraphrase. I stopped at the email signup form (Work or school email, Password, Name) and did not create an account.

Not tested

To be clear about what this session did not evaluate:

• I did not test the Humanize feature or its 6 uses/day allowance.
• I did not test the Grammar Checker, Summarizer, AI Detector, plagiarism checker, browser extension, or file export/download features.
• I did not paste a text longer than 125 words to observe the exact interface cutoff message.
• I did not complete account creation, email verification, password login, Google/Facebook/Apple SSO, or any paid checkout/card flow.
• I did not test document persistence across sessions, document history, file upload, collaboration tools, API access, or data deletion policies.
• I did not test other Premium-only modes (such as Academic, Simple, or Creative) beyond hitting the Formal mode gate.
• I did not record stopwatch-level latency measurements.

No assumptions

• I do not assume the 125-word cap translates into a specific character limit; QuillBot explicitly measured and displayed words.
• I do not assume Humanize or any other tool listed on the pricing page is usable without restrictions simply because it appears in the menu.
• I do not assume the rewritten text is grammatically or stylistically superior across every domain; I only evaluated the specific project-tracker output received.
• I do not assume the email signup supports passwordless magic links; the form explicitly contained a password field.
• I do not assume the direct /paraphraser URL is broken globally on every browser; it failed in this specific Chrome session while the sidebar link worked.

Who it's for

I’d reach for the free tier when I need a short email rewrite, a Slack update tightened, or one clunky paragraph cleaned up — without creating an account.

I would not try to drop a multi-page report into the free box, and Formal (or other Premium tones) stayed locked until Upgrade.

Verdict

For a cold visit, QuillBot still let me rewrite under the 125-word free cap with Standard and Fluency, no credentials required. The word ceiling and locked Formal mode are real limits. For short unauthenticated paragraph fixes, the free paraphraser did what I came for.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/quillbot (official website, verified during review).

What I actually did

I opened the homepage anonymously on 7 September 2026. No cookie banner. The search box sat front and center under “Explore academic papers in a visual graph.”

I typed Attention Is All You Need and picked the matching suggestion. The graph showed up in about 2 to 3 seconds. No email. No password. No mandatory registration. I honestly expected a soft gate after the first build. It did not show up.

A first-use overlay, “How to read the graph,” sat on top of the canvas. Helpful once. Awkward as a first impression, because it covers the thing you came to click. I closed it. The origin node was Attention is All You Need (Vaswani et al., 2017), with about 190,520 citations at the time of the test. Node size tracked citation scale, and a year legend ran from older work into recent years.

I clicked related nodes, including LLaMA: Open and Efficient Foundation Language Models and Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation. Each click filled a right-hand detail pane with authors, year, citation count, external-open icons, and an abstract. That pane is where I actually read. The graph labels themselves overlap too much for long titles.

Then I opened Prior works and Derivative works. Both were usable anonymous tables with sortable columns and a Download control. Prior works showed older anchors like Effective Approaches to…, Neural Machine Translation…, Dropout…, and ImageNet. Derivative works showed later papers such as Open Problems in Applied Deep…, Neural Machine Translation…, and Deep Representation…. Those tables are why I would come back.

I changed the view two ways. I switched to List view (URL went to /list). In filters, the panel reported 13 Open access, 32 PDF available, and 0 Code available. I turned on Open access. The toolbar showed Filters (1), the left list narrowed, and non-matching nodes on the graph washed out. Readable, not broken.

More → Download worked without login. I got ConnectedPapers-for-Attention-is-All-you-Need.bib at about 57.5 KB. Share opened a free modal (Facebook, LinkedIn, Twitter, Email, Copy link, plus a checked “Link to this graph” option). No account wall.

Save was the wall. It sent me to /auth with Log in / Sign up, email, Google, and institution options. I did not create an account. The graph work I came for was already done anonymously.

Pricing (Individuals, annual billing): Free USD 0 with 5 graphs per month; Academic USD 6/month; Business USD 20/month. Academic and Business claim unlimited graphs. The page lists similarity graphs, prior/derivative works, multi-origin graphs, saved papers, and graph history on the plans. I entered no card.

What worked and what annoyed me

The speed is the point. A full neighborhood in a few seconds, no email, is rare for research tools. Citation-sized nodes plus a year legend make browsing feel like a map, not another search results page.

Prior works and Derivative works are the sleeper hit. Sortable citation and year columns, plus anonymous download, beat hunting through reference lists by hand. BibTeX export and Share both worked without login. That boundary is clear: explore and export free, Save needs an account.

What kept it from a 5/5 were the overlapping node labels in dense clusters, the blocking first-use overlay on top of the main canvas, and Save sitting behind auth. The Open access filter also fades the graph instead of rearranging it. Fine once you know. Still a bit washed-out.

Not tested

• No email signup, verification, password, Google login, or institution login
• No second-origin graph for BERT (I changed the view with List and Open access instead)
• No PDF/DOI/publisher full-text opens, graph history, multi-origin add, or export formats beyond the observed .bib
• No payment, upgrade checkout, or credit card flow
• I also did not burn through the free “5 graphs per month” cap, so I cannot say how the limit feels on day 20 of a lit review

Who this is for

If you need to survey a subfield around one key paper, click into neighbors, skim Prior/Derivative tables, and grab a BibTeX file in a few minutes without handing over an email, this session did that. If you need saved graphs, history, or more than five graphs a month on the free plan, you are looking at Academic or Business. I did not run those paid paths, so I will not score them.

Verdict

Connected Papers gave me a real citation map before any account wall. Graph build was fast, Prior/Derivative tables were useful, and anonymous BibTeX plus Share worked. Crowded labels and the Save gate keep it under a 5.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/connected-papers (official website, verified during review).

On ToolifyNav, Uizard is listed as a freemium design and prototyping tool (official site: uizard.io). I went directly to the web app, set up a standard free account, and worked through the entire path from prompt to preview.

No paid trial, no card details entered. Everything here is strictly what a free user sees in one sitting.

The Hotmail Junk folder detour

I started at https://app.uizard.io/sign-up/ using a standard Hotmail email and password. I intentionally skipped Google and Microsoft single sign-on because I wanted to see how the regular email verification behaved.

The signup itself was fine. The weird part was waiting for the email. The verification message from noreply@uizard.io never showed up in my inbox. After a minute of staring at an empty inbox, I checked my Outlook/Hotmail Junk folder, and there it was. If someone doesn't think to dig through spam, their onboarding ends right there.

Once I clicked the verification link, I cleared the required onboarding questions and accepted the terms. I declined the optional marketing consent, and the app routed me to the workspace through https://app.uizard.io/login.

Prompts to screens: Bean & Brew in 45 seconds

Inside the workspace, I selected the free Autodesigner text-to-design option.

I typed in a prompt describing a mobile login flow for a coffee shop concept called Bean & Brew. I hit generate and waited.

The build took roughly 45 seconds. I honestly expected a single disconnected mockup screen. Instead, I liked what came out more than I anticipated: Autodesigner generated a full multi-screen prototype with a welcome screen, login and registration flows, plus menu and ordering screens.

The editor laid out thumbnail previews cleanly. Up in the interface header, a small counter showed 163/400 components used.

The first result was better organized than I expected. Form fields lined up, the coffee theme carried across screens, and the pieces felt like real UI components rather than flattened image blocks.

Editing on the canvas and catching a text glitch

I wanted to see whether the generated screens were actually editable, rather than polished mockups I couldn’t do much with.

I clicked into the cover/login screen to test the vector editing. The design included an orange circular-arrow button icon. I selected it and swapped the fill to a light cream color (roughly #f2e0ce). The canvas updated instantly, and the surrounding layout stayed perfectly in place without distortion.

Then I clicked Preview to test the flow. The prototype launched right inside the browser without throwing up an upgrade wall. Tapping through the screens felt responsive and fluid.

Then I spotted a careless string bug right on the header: Welcome back â© sign in.

It is a minor string parsing bug, not a product-breaking failure. Still, on a design tool that sells visual polish, a broken symbol jumps out immediately.

Sharing and the Export-to-Handoff surprise

Once the screens were in decent shape, I looked into sharing and exporting the project.

Clicking Share opened a modal titled “Share Bean & Brew Login” with three tabs: Share, Publish, and Embed. Under the Share tab, the text stated: “Add new members to your project” and “People invited will be added to your organization.” It also pointed out: “This project is Public to your organization, everyone within the organization can access.”

So sharing defaults to organization members rather than spitting out a public link on the spot. I didn't send external invites for this run.

Exporting was more disorienting. I clicked Export in the top right, expecting a clean little download dialog to grab a JPG, PNG, or PDF. Instead, the screen shifted into Handoff Mode with an intro banner: “Welcome to Handoff Mode 👋 Select any screen or component to export visuals and inspect React and CSS code.”

For a second I thought I clicked wrong. But developer handoff and code inspection are paid features, so I stopped there. I didn't trigger a paid trial or download paid assets, and I couldn't confirm a standalone direct image download dialog outside of that Handoff view.

Free plan limits vs. paid tiers

I pulled up the pricing page on uizard.io to see what constraints were governing the session:

• Free ($0): “For students and hobbyists.” Unlimited free viewers and commenters, 3 AI generations per month, Autodesigner 1.5, 2 projects, and 10 free templates. The feature list specifies up to 5 screens per project, 400 components per organization, up to 5 saved custom templates, self-serve support, and Export to JPG, PNG, and PDF at 1x resolution only.
• Pro ($12/month billed annually, marked “Save 40%”): 500 AI generations per month, Autodesigner 2.0, developer handoff in React and CSS, up to 100 projects, private projects, and all templates.
• Business ($39/month billed annually): 5,000 AI generations per month, faster AI generation, custom brand kits, unlimited projects, and priority support.
• Enterprise: Custom contract pricing with unlimited AI generations, unlimited teams, design-system setup, and an AI data SLA.

The 400-component limit sounded generous until I noticed this single test project had already used 163. I went back to the pricing page because that didn’t match how “free forever sandbox” usually feels in my head. Three AI generations per month and 400 components disappear faster than they sound. With 1x-only exports on top, I wouldn’t count on the free plan for ongoing work. The limits start to matter pretty quickly.

Who this actually works for

If you are a founder or product lead who needs to take a concept from a rough idea to a clickable, editable multi-screen prototype in under a minute, Uizard does the job. That 45-second Autodesigner run delivered an actual working wireframe layout, not a flat JPEG.

I wouldn’t count on the free plan for ongoing work, though. Three AI generations per month and 400 components disappear faster than they sound.

Verdict

Uizard generated a coherent, multi-screen coffee app prototype in roughly 45 seconds on its free tier. The canvas lets you make quick vector tweaks, and Preview works without shoving you into a paywall first.

What kept it from a 5/5 were the smaller annoyances: the verification email went to Junk, Export unexpectedly pushed me into Handoff Mode, the free limits are tight, and Preview showed a broken text string.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/uizard (official website, verified during review).

That is my bias from the start.

Most AI photo editors bury the actual work behind a funnel. I wanted to know if Photoroom would let me do one job without an account. If the public widget worked, great. If it shoved me into a login before a single export, I would say so.

On ToolifyNav, Photoroom is listed as a freemium Image Generation & Editing tool. Site: photoroom.com. Pricing on the listing: “Free tier / paid Pro subscription and API credits.” I treated that as a map, not as proof.

I ran two sessions on 2 September 2026. First I stayed logged out on the public background remover. Later the same day I signed in with email, still on the free plan, in the default personal space. That second session does not make this a full-product review. It is what I actually clicked.

Session 1: logged out on the public widget

I stayed logged out on the Photoroom background remover. No Google. No Facebook. No Apple. No email.

I uploaded a simple phone product photo. A silver Samsung on a desk. Nothing fancy. No jewelry. No busy flat-lay. Just one product.

The widget processed it in about 5 to 10 seconds.

I got a clean transparent cutout. Edges looked tight around the phone against the checkerboard. No watermark on that result. No login prompt. No credit counter in that widget.

Download worked in the browser.

That part is good.

I will not dress it up. I could not inspect the downloaded file on disk. So I will not claim export resolution, file format, or alpha quality at 100% zoom. I saw a clean cutout on screen, and the browser let me download it. That is all I can stand behind.

After download, I tried the obvious next step. A cutout without a new background is only half the job I came for.

Background swatches did nothing.

Before-After did nothing.

Add background did nothing.

I clicked them. The widget did not change the result. From a logged-out session on 2 September 2026, those controls did not work. I would have accepted a paywall. I got dead buttons instead. That is worse than a paywall. A paywall is at least a sentence.

I also tried the sample thumbnails sitting next to my result. They failed.

The error was blunt: “Oh no, something went wrong! Please try again later.”

Then I uploaded a more complex flat-lay. Same error. Same dead end.

So the one logged-out success I had was a simple product shot I chose myself. The samples and the harder image did not go through. That is a bad ratio for a public demo.

Session 2: email sign-in, same day

The full app at app.photoroom.com/create is behind signup. Google, Facebook, Apple, or email.

Later that same day I used email sign-in. Photoroom does not set an account password for email. It is a passwordless magic code or link. I landed in a free personal space. I did not start a trial. I did not enter a card.

Batch, Brand Kit, Virtual Model, and the API were visible in the sidebar. I did not open them. I am not going to review them.

Cutouts are free once you are inside

I uploaded a product photo in the logged-in editor. The background came off automatically.

Edit Cutout was there. Erase and Restore were there. Guided versus Manual was there. Cutout Suggestions were there. All of that ran on the free space. No upgrade prompt on those controls.

That is the job the public widget only half-finished. Signing in did not make the cutout smarter. It made the editor usable.

Generative AI waited, then asked for Pro

I tried a generative AI background. I used “Describe an edit.” Generate spun for about 20 seconds. Then the UI said “Upgrade to Pro to unlock this feature.”

No output. No image. The paywall hit after the wait, not before.

That is a worse pattern than a locked button. I sat through the spinner as if work was happening. It was not. If a feature is Pro-only, say so before I wait.

Usage is a modal, not a page

I looked for usage. Photoroom’s /usage URL 404s. Usage is a modal.

The modal showed AI credits 0% used. Exports 0/100 used. It renews October 2. The same modal said “Get Pro to unlock more.” Upgrade Space to Pro banners sat on the screen and did not go away.

I had used none of those credits yet. The pitch was still there. So the free space is a cutout editor with a 100-export cap and a persistent upgrade banner. It is not a quiet studio.

Pricing as shown on the site

I read photoroom.com/pricing in USD. I did not start a trial. I did not enter a card.

Yearly, billed monthly: Pro $7.50 (was $12.99), Max $20.99 (was $34.99, marked most popular), Ultra $82.50 (was $99).

Monthly: $12.99 / $34.99 / $99.

Pro listed 8000 AI credits and 1000 exports. The CTAs said “Start free trial.” I am not repeating vendor sales-uplift stats as if I verified them. Marketing numbers are not test results.

The only usage counter I actually saw was that free-space modal: 0% AI credits, 0/100 exports, renews October 2.

Who this is for, based on what I saw

If you have a simple product photo and you only need a transparent cutout, both the public widget and the free logged-in editor did that job. Fast. Clean on screen. After sign-in, Edit Cutout with Erase, Restore, Guided, Manual, and Cutout Suggestions ran without a paywall.

If you need new backgrounds from the same logged-out screen, I cannot recommend it. The controls sat there and did nothing.

If you need generative AI backgrounds or “Describe an edit” on the free space, I also cannot help you. Generate spun for about 20 seconds, then asked me to upgrade. No output.

Do not treat this as a review of Batch, Brand Kit, Virtual Model, or the API. Those items were in the sidebar. I did not use them.

Verdict

Photoroom can cut a simple phone product photo in a few seconds, logged out or logged in. That is the honest win.

The logged-out widget then failed me. Background tools did nothing. Samples and a complex flat-lay threw “Oh no, something went wrong!” The full app asked for an account.

After email sign-in, the free editor did the cutout work the widget would not. Generative AI still did not. It waited, then paywalled. Usage lives in a modal, not at /usage, and the free space already pitches Pro at 0% used.

I like tools that finish the job they start. This one starts the cutout. It does not finish a generated background on the free path I used.

Final rating: 3 out of 5.

The cutout worked. A 20-second spinner that ends in “Upgrade to Pro” is not a higher score. Signing in did not earn Pro features I still did not get.

Sources: https://toolifynav.com/item/photoroom (official website, verified during review).

What the product actually is

The pitch is narrow on purpose. You get a throwaway address, an inbox that receives messages, and a copy control. That is the whole job. The official site also advertises multiple selectable domains, which matters in practice: when a signup form rejects a well-known disposable domain, switching to another one is usually faster than arguing with the form.

Next to the inbox the site bundles a few small utilities — a password generator, an email validator, an IP lookup and a JSON formatter among them. It is not a developer toolbox, but it fits the same short errand as the inbox itself.

Where a temporary inbox is the right tool

Use a disposable address when the cost of the address leaking later is close to zero: a one-time download gate, a free trial you will not renew, a newsletter check, or a test account for a signup flow you are auditing. In those cases a temporary inbox keeps the follow-up marketing away from your real address.

The value is containment, not anonymity. The address expires and the messages go with it, which is the feature and the risk at the same time.

Where the trust ceiling sits

A temporary inbox is not private in the way a personal inbox is. The address is generated for you rather than tied to your identity, and anyone who learns it can potentially read what arrives. That makes it a poor fit for anything you would have to recover later: banking, work accounts, password resets on accounts you keep, client correspondence, contracts, or a purchase you might need a receipt for.

Treat it as a disposable raincoat, not a security system. It keeps your real inbox dry during a short errand. It does not make you invisible: the sites you visit can still see your IP address, browser and cookies.

Two questions decide whether it is safe for a given errand, and the product pages should answer them plainly: how long the address lasts, and how long received messages stay. If those numbers are not on the page you land on, ask support before you point a real workflow at it.

What this edition did not verify

This assessment is based on FreeTempMail’s public pages and the ToolifyNav directory listing, not on a live inbox session. That boundary matters, so here is exactly what sits outside it:

• No live inbox run: no address was generated, no message delivery was timed, and no verification link or code was received end to end.
• Domain behaviour unverified: which domains are offered at any given moment, and which signup forms accept or reject them, was not tested.
• Lifetime numbers unverified: address expiry and message retention were not measured, because they are not published in a form that could be measured.
• Utilities unverified: the bundled password generator, email validator, IP lookup and JSON formatter were not run.
• Nothing about deliverability, uptime, logging or retention policy was confirmed with the vendor.

Who it suits

For throwaway tasks — trial signups, one-time downloads, newsletter checks and test accounts — FreeTempMail fits the job without asking for anything first. That low friction is the part worth keeping.

For anything you would need next month, use your real address or a mailbox you control. That is not a criticism of this product; it is the correct way to use any disposable inbox.

Verdict

FreeTempMail does the basic job with no registration and no funnel, and the bundled utilities are a sensible add-on for the same errand. The category’s trust ceiling, plus the fact that this edition is documentation-level rather than hands-on, is why it stays a 4 rather than higher.

Final rating: 4 out of 5.

Sources: https://toolifynav.com/item/freetempmail (official website, verified during review).