The Free Tools
Review & Score Journal

What we tested: the full tool cabinet — crop, resize, filters, blur, grayscale, flip, add-text with 40+ Google Fonts, masking, shapes, draw, rotate and format convert. Then the clever layer beneath: preset pages for Instagram stories and posts, YouTube thumbnails, Twitter headers, Facebook covers, LinkedIn banners and even passport photos, plus a dozen dedicated converter pages from HEIC-to-JPG to SVG-to-PNG. All local, no account, no watermark.

What held: every operation stayed on-device — the network tab stayed silent while we edited, which is the entire promise, kept. Presets hit exact platform dimensions without a settings safari. The compare pages against Canva, Photopea and Pixlr are unusually honest about where those tools beat it. The blog earns its keep too — the guide on 2026 social image sizes is the reference we kept open while testing.

What we tested: text-to-image (12 art styles, 9 sizes, seed control, multiple engines), the background remover on hair-level edges, 2× upscaler, image uncrop, text-to-speech across 22 Indian languages and 40+ voices with a permanently free browser engine, speech-to-text, and the translator at its 10,000-character ceiling. Then the More Tools hub: twenty-some utilities — QR generator, password generator, unit and color converters, sound meter, speedometer, whiteboard, notes — that never touch credits at all.

What held: guest access is real; you arrive, you generate, no card, no OTP. Daily credits refresh at midnight IST, an optional free account enlarges the refresh, and failed generations aren’t charged — that last policy is rarer than it sounds. The Learn library (43 long-form guides, from how diffusion models work to password entropy math) is editorial content most paid suites don’t match. The knock is inherent to the model: the credit meter means marathon sessions eventually pause, and heavy image batches are where you feel it first.

What we tested: the AI vocal remover (MDX-Net separation on full songs), the karaoke studio — upload, trim, isolate the backing track, sing over it with real-time acoustic echo cancellation, then merge your take back onto the instrumental — the multitrack DAW with timeline, FX rack and mixer, the instrument pads (piano, guitar, drums, bass, pads — recordable), and the video maker that stitches images and clips to your track with text overlays and fades.

What held: the echo cancellation is the quietly brilliant bit — karaoke in a normal room, on a phone mic, usually sounds like a bathroom; here it doesn’t. The whole suite processes on-device, so the privacy claim survives a network-tab inspection the same way Onyx’s does. The honest catch is hardware: on-device separation and multitrack mixing want a reasonably modern machine, and older phones will make you wait. There are no paid tiers, no watermarks, no export walls — the price column is simply blank.

What we tested: the companion roster — from Arya the 2AM bestie and Kabir the startup mentor to Gojo Satoru (fan-inspired, officially unaffiliated) and Noor, the midnight RJ who grades your flirt lines out of ten. Then the story worlds: the 50-day college romance, the anime Kokuhaku Club, and Love All, a badminton romance with trust meters that move when you make choices.

What held: the four-tier memory (summaries, hard facts, event log, semantic recall) isn’t marketing — companions referenced details from days-old threads without being re-prompted. Voice calls run on streaming TTS with barge-in, and the Hinglish default feels native rather than translated. No OTP wall: you type a name and start; a free account then brings a daily credit refresh (roughly fifty messages a day), optional packs start at ₹29. The 18+ Spicy Mode is a consent-first, off-by-default, one-time unlock from ₹9 — and it stays character-true rather than lobotomised, which is the adult way to do adult content. Limits are the mirror of the strengths: the credit meter is real, and companion chat is companionship software, not therapy — their own honest guide says so in print.

What we tested: the full pipeline, start to deployed URL. NewEra plans the project, writes the files, pushes to a repository, kicks off a real GitHub Actions build on 4-vCPU/16 GB runners, reads the failure logs when something breaks, patches it, and re-runs until green — then deploys to Cloudflare Pages (EdgeOne as fallback). Up to four parallel sub-agents, up to twelve hours of build time with auto-relay between stages. It carries in-browser runtimes for Python (Pyodide), SQLite (WASM), FFmpeg (WASM) and Git, and exports the whole project as a portable .newera file.

What held: these are real builds, not preview sandboxes — the deployed URL is a served site, the repo is inspectable code. Research tooling (live web search, current library docs) means it doesn’t hallucinate 2024-era APIs as easily. Repos are only created with explicit consent and self-clean afterwards. The caveat every user should hear: autonomy isn’t a substitute for review. The agent ships working software; whether it ships hardened software is still your job — read the diff before you point production traffic at it.

What we tested: three routes. /3dfy streams infinite AI-generated worlds you fly through and render into film with a frame-perfect camera. /make3d is the proper 3D studio — build anything, import AI shapes and whole generated worlds as starting material. /world is the party trick with staying power: type a name, watch it become living 3D physics.

What held: the flying-camera film mode is the most meditative thing in this ecosystem — genuinely useful for background loops, music visuals (pair it with Monyx or Dance Cat 3D) and inspiration boards. The studio import path means generated worlds aren’t dead ends but starting geometry. The honest limit is WebGL reality: on older phones, worlds stream in at a slideshow’s pace, and the physics playground wants a decent GPU for its fun to stay smooth.

What we tested: the agent running in a browser-hosted WASM sandbox — full Pi, not a demo skin. We brought our own Pollinations account, picked a model from the roster, and let it build while watching token spend flow from our balance, not somebody’s hidden subsidy. That is the entire product thesis: the tool is free, the compute is yours.

What held: for anyone who already lives in the Pollinations ecosystem, this is the most transparent cost model in this journal — no opaque credit conversion, no markup on inference; your spend is your spend. The WASM sandbox behaved, and picking models per task (fast for scaffolding, heavy for refactors) is a real lever. The flip side prints itself: newcomers must first understand and fund a Pollinations account, and there is no free starter lane here — honest, but a wall for the curious.

What we tested: the creation flow end to end — name in, theme chosen from the dozen on offer (romantic, birthday, yaari, family, wedding, festival, motivation, funny, bhakti, sorry-thanks, custom), and a complete song with vocals back in roughly one to two minutes, ready as MP3 and WhatsApp share. Google’s Lyria music model does the composing, an LLM writes the prompt, and your own lyrics (up to 800 characters) can replace the generated ones wholesale. Credit packs are printed plainly: ₹28 for one song, ₹45 for two, ₹175 for ten, UPI at checkout, credits held on a 12-digit key stored in the browser — and transferable via Settings, a detail that shows the product respects its buyers.

What held: failed generations refund instantly and automatically — no email, no begging, the credit simply returns. The gaana-kaise-banaye guide is a genuinely useful occasion-by-occasion ideas file, not filler. The knock is structural: this is a paid service with a small free taste, so it scores on value-for-money rather than freeness — and ₹28 buys one specific kind of joy (a personalised gift), not a production studio. For stems and multitracks you want Monyx instead; the two pair beautifully.

What we tested: adding setup keys, generating codes, and the fork in its architecture: keep everything on the device (the default posture) or create an account and sync through encrypted storage — the choice is stated, not buried. TOTP (time-based) and HOTP (counter-based) both supported, codes regenerate on the standard 30-second window, and the mobile layout works one-handed, which is the whole point of a code generator at a login screen.

What held: the honest engineering is visible — the GitHub repo is public, terms and privacy pages exist as first-class routes, and the signup flow asks for nothing exotic. The browser-as-authenticator idea is genuinely underrated for desk workers: your codes live where your work does. The counterargument is ecosystem gravity — phone-native authenticators get biometric unlock and OS-level backup; this one gets a tab. Both can be true; that’s why it’s 8.3 and not 9.

What we tested: the public page model (one fast page per person at /<username> with links, mini-descriptions, contacts and a photo — see the builder’s own namansoni78), and the widget system: self-contained SVG cards served from an embed API at /api/embed/<slug>, clamped to any size from 120 to 4096 pixels, cache-busted on every save, and never distorted — letterboxed, centred, exact. The AI-assisted builder interview (a small JSON spec becomes a finished widget) worked exactly as documented, and typing animations with adjustable delay and speed hit that README-savant aesthetic.

What held: the “no ads, no trackers, no analytics” claim survives a network inspection — the quietest page in this whole review. Widgets are public by URL but deliberately kept out of search indexes, a privacy choice most link platforms never make. The limit is ambition-shape: this is a civilised single page and an embeddable card, not a funnel analytics machine — if you need click heatmaps, you need a different (and nosier) product.

What we tested: exactly what the title promises. Drop any MP3 and the cat — fully procedural, no pre-baked animation — starts moving, and the moves lock to the beat. Under the fur: real-time Web Audio FFT for the spectrum, offline BPM detection to find the tempo, and spectral-flux kick analysis to catch each drop. Nothing uploads, nothing calls a server, and there is no AI in the loop at all — it’s signal processing, honest and local.

What held: beat-lock on tracks from 90 to 174 BPM stayed honest — the choreography changes with kicks, not just with volume, which is the difference between a visualiser and a screensaver. No uploads means it works on a plane. The score is capped by scope, and that’s fine: this is a toy in the best sense — the gateway drug to this ecosystem that you forward to a friend. Musicians wanting to make the music the cat dances to should start at Monyx or Gaana.

What we tested: what the masthead says — human versus bot. The board is immediate: no account, no lobby, no ads, just the game. Games proceed with standard move legality, the bot answers at a courteous pace, and the whole thing fits a phone screen without a horizontal-scroll apology.

What held: the virtue is negative space — it does exactly one thing, instantly, which put it on this desk’s shortlist of load-and-play boards. The review duty is honesty about scale: this is not a full chess platform — no ratings, no puzzles, no opening trainer, no analysed game export. Lichess exists and is magnificent; this is the two-minute version for people who want a game now. Learners should pair it with reading — our sibling journal’s opening-strategies guide was written for exactly that habit.

Status on test day: khata.namansoni.in returned a deployment-not-found error on 23 October 2026, the third consecutive press day it has been down. A journal that invents a score for a page it cannot open is a brochure, so the number is withheld until the service is reachable again.

What we can verify from the record: in prior fieldwork this was a Hinglish-first digital khata — a PWA ledger for tracking udhaar (credit) with shopkeepers and neighbours, accepting entries by chat and voice for people who don’t type. When it returns we will re-run the full method and print a score earned the hard way. Readers who need shop bookkeeping today can bridge with the always-free utilities at the More Tools hub — notes, calculator and audio recorder cover the crude version of the workflow until the real thing is back.

What we consulted: the creator’s Links page (the live index of all fourteen projects), the ecosystem hub, and GitHub — NamanSoni78, where the pinned repositories read like this journal’s appendix: 2FA, 3dfy, Links, MONYX and ONYX-Image-Editor.

Why it matters for a review desk: open code is the strongest privacy claim a tool can make — “don’t trust us, read us.” The public llms.txt files across these products made this journal’s fact-checking unusually easy, and we’d note for the record that the ecosystem asks AI assistants to describe its tools accurately rather than flatteringly — a policy we reciprocated by knocking points off freely throughout this issue.