How it works

The simple version

When you drop an image into Compress, your browser opens it directly. It never travels across the internet, and nothing is sent to our servers.

We then re-save the image using a tighter recipe — the kind of careful repacking that makes a file much smaller without making it look obviously worse. You pick the quality and the output format, and your browser does the rest. When it's done, you download the smaller file straight from your own device.

Under the hood

Compression runs on a set of open-source codecs ported from the Squoosh project to WebAssembly. They ship in the jSquash family of packages:

• @jsquash/jpeg

  Wraps MozJPEG, Mozilla's tuned JPEG encoder. Lossy. The quality slider maps directly to MozJPEG's 1–100 scale; 75 is a good default for photographs.

• @jsquash/png

  Wraps oxipng, a Rust-based lossless optimizer. PNG output is rewritten with the smallest matching deflate stream. The quality slider does not apply here — PNG is always lossless.

• @jsquash/webp

  Wraps Google's libwebp encoder. Lossy. Produces files around 25% smaller than equivalent-quality JPEG for most photos.

• @jsquash/avif

  Wraps libavif from the Alliance for Open Media. AV1 still encoder, producing the smallest files of the four formats. Slowest to encode — marked 'experimental' in the UI for that reason.

Each codec is a WebAssembly module — compiled code running at near-native speed inside your browser. We load all four inside a dedicated Web Worker, so the encoding work happens off the main thread. That's why the page stays responsive when you compress a batch of large photos.

Nothing leaves the worker except the compressed bytes, which are wrapped into a Blob and offered to you as a download. No network request carries your image data at any point in this flow.