I wanted Squoosh with batch and a KB budget, so I built a client-side compressor

I compress images for websites, side projects, and client work. For years that meant TinyPNG or Squoosh. Neither matched how I actually work.

TinyPNG is fast and familiar. It also wants the file. You upload. The free web tool caps you at 20 images per session and 5 MB each. After that you pay, wait, or split a folder into batches of twenty. The compression is fine. The trust model is “someone else’s server, hope they delete it.” I did not want that for client photos, UI screenshots, or anything I would not put on a public CDN first.

Squoosh is the other extreme, and I still use it for one picture. Drop a file, pick MozJPEG / WebP / AVIF / JXL / Oxipng, compare sides, download. The image never leaves the browser. That part is right. The workflow is not: one file at a time, no folder drop, no ZIP, no “make this under 100 KB.” The CLI and library were deprecated years ago. The web app still works; it is not where I go when a public/images folder lands on my desk.

I am Mehdi Tareghi. I build software from Iran. I got tired of the gap and built Asset Melt: Squoosh-grade WASM codecs, plus the batch and budget pieces I kept missing.

What it does

Studio is a static app. Open it, drop files, encode on your machine, download. No account.

Codecs run in Web Workers via WebAssembly: MozJPEG, AVIF, WebP, Oxipng, and JPEG XL (plus QOI). HEIC from a camera roll decodes locally with libheif. PNG is lossless Oxipng by default, with optional palette reduction when you want lossy PNG-8.

The parts I built because Squoosh would not:

• Batch and folders. Drop a directory. Nested images are walked; non-images are skipped. The queue is reorderable. You can add files while a batch is encoding, pause mid-run, and export a ZIP that keeps relative paths (products/a/hero.webp). There is no 20-image cap. Device memory is the cap.
• A KB budget. “Quality 75” is not a size. I often need under 50 KB, under 100 KB, or under 200 KB for an OG image, a newsletter hero, or a card. Studio binary-searches quality (and PNG palette size) against that cap for JPEG, WebP, AVIF, and JXL.
• Platform kits. Fit-to-size presets for OG, X, Instagram, YouTube, favicon, App Store screenshots, and newsletter widths — as ZIPs, not a scavenger hunt through a resize panel.

You can also export more than one format in a single run (AVIF / WebP / optional JPEG). Recipes live in the URL (?recipe=); photos do not. Code is at github.com/mehditareghi/assetmelt.

Images stay on the device. The site is not invisible.

Image bytes never upload. There is no image API. I cannot see, store, or recover your files. Close the tab and in-progress work is gone unless you exported. That is architecture, not a footer promise.

The site still has UI analytics. I use Google Analytics 4 for page views and coarse events (files added, processing finished, export — counts and formats, not filenames or pixels). Sentry gets crashes, traces, and sampled session replay of the website chrome, not your pictures: media is blocked, Studio previews are local blob URLs. Vercel records web vitals. It is all on assetmelt.com/privacy. If a privacy pitch hides telemetry, it is a lie. This one does not.

Try it in about 20 seconds

1. Open assetmelt.com/studio. No signup.
2. Drop a JPEG, a PNG, or a whole folder.
3. Pick WebP or AVIF, or turn on a size budget — under 100 KB is a good first test.
4. Process (Cmd/Ctrl+Enter).
5. Download the file, or ZIP the queue.

If you already know the cap, these open Studio with the budget on: under 50 KB, under 100 KB, under 200 KB.

First load pulls WASM modules. After that, work is local CPU. A cheap laptop will feel a large AVIF batch. That is expected.

Squoosh vs TinyPNG vs Asset Melt

Features only. No invented traffic numbers, no “X% smaller” claims.

	Squoosh	TinyPNG (free web)	Asset Melt
Where encoding runs	Your browser (WASM)	Tinify’s servers (upload)	Your browser (WASM)
Account	No	Not required for the free web tool	No
Batch	One image at a time	Up to 20 images / session, 5 MB each	Folder drop, queue, ZIP; RAM is the limit
Hit a KB target	Quality / codec sliders	Automatic compression, no “under N KB” budget	Binary-search under 50 / 100 / 200 KB
Codecs / formats	MozJPEG, WebP, AVIF, JXL, Oxipng, QOI, WP2	JPEG, PNG, WebP, AVIF	MozJPEG, WebP, AVIF, Oxipng, JXL, QOI; HEIC decode
Platform kits (OG, X, App Store, …)	No	Resize via paid / API, not kit ZIPs	Yes
Open source	Yes	No	Yes
Images uploaded	No	Yes	No
Site analytics	Yes (Google Analytics on the app)	Yes (it is a hosted service)	Yes — UI only, disclosed on /privacy

TinyPNG is still the right choice if you want a hosted API, a WordPress plugin, or you are fine uploading and staying inside the free 20. Squoosh is still the right choice if you want a side-by-side codec lab for one file. I use both as references. I built Asset Melt for the days I have a folder, a size cap, and no reason to send pixels off-device.

I did not write MozJPEG. I wired the same family of engines Squoosh proved in the browser (@jsquash, libheif) into a React app, then added a queue, a budget search, and the export paths I kept wishing existed. There is no cloud history and no team workspace — there is no backend that receives photos. It is free. No plan page.

If that is the workflow you already have in your head — batch, a KB number, files that stay put — open the studio and drop something.