A single-file, fully client-side background remover. Drop in a photo, the AI segments the subject, and you download a transparent PNG. No server, no uploads — your image never leaves your machine.
Open index.html in any modern browser (Chrome, Edge, Firefox, Safari).
That's it — there is no build step and no install.
First run downloads the AI model (~tens of MB) from the jsDelivr CDN
(@imgly/background-removal + its ONNX segmentation model), so you need an
internet connection the first time. The browser caches the model, so later
runs are much faster.
- Drag & drop, file picker, or paste from clipboard (PNG / JPG / WebP)
- Real progress reporting while the model downloads and runs
- Original vs. result toggle on a checkerboard transparency preview
- One-click transparent PNG download
- Large images are downscaled to max 2048 px on the long edge before processing
- Offline fallback mode: if the CDN or model can't load (e.g. you're offline), a pure-canvas color eraser is offered — click a background color, tune the tolerance slider, with contiguous (flood-fill) or global erase, undo, and restore. Works with zero network access.
- Internet is required on first run for the AI path (model comes from a CDN).
- Output resolution is capped at 2048 px on the long edge.
- The AI model struggles with very busy scenes, fine hair/fur, and semi-transparent objects, like any segmentation model.
- The offline fallback is a simple color-threshold tool — it works best on solid or near-solid backgrounds, not complex ones.
- Inference runs on your device; very old hardware may be slow. Some browsers in private/incognito mode won't cache the model between sessions.
MVP.