Files stay on your device. Enter send · Esc close drawers
Remove background, keep subject. PNG with alpha.
Background removal isolates the subject by predicting a per-pixel alpha mask. Cutout (transparent PNG) runs the standard U²-Net variant — a compact (~5 MB) salient-object segmentation model that handles hair, fur, fabric edges, and translucent materials better than chroma-key approaches. The mask is composited back onto the source as a transparent PNG, ready for product photography, social posts, design comps, or photo-ID submissions where a clean white background is required. Everything runs locally in WebAssembly — no network upload, which matters when the input contains people or proprietary product shots. The preset outputs Transparent PNG.
| Processing | 100% in your browser — no upload |
|---|---|
| Cost | Free, unlimited |
This preset is a deterministic recipe — every step has fixed parameters so the result is byte-identical across runs of the same input.
The standard background-removal model is a quantised U²-Net checkpoint, around 5 MB on disk. It runs in your browser via ONNX Runtime Web — the WebAssembly backend is the default; WebGPU is used opportunistically when available. The model produces a per-pixel alpha matte rather than a binary mask, which preserves hair, fur, and translucent edges more cleanly than chroma-key approaches.
First-run time on a 1024×1024 input is around 1-2 seconds after the weights are cached; subsequent runs hit the cached model instantly. Like all browser-side processing here, the source image is never uploaded.
A quantised U²-Net checkpoint tuned for standard subjects. The model produces a per-pixel alpha matte rather than a binary mask, which preserves hair, fur, and translucent materials more cleanly than chroma-key approaches. Total download is around 5 MB and is cached after first load.
A transparent PNG. The original RGB pixels are preserved where the model is confident the subject exists; the alpha channel ramps from 1.0 (subject) to 0.0 (background). Drop the result onto any background — Mochi's preview shows it on a checkerboard so you can see edge quality.
Yes. The U²-Net checkpoint runs in WebAssembly via ONNX Runtime — no upload, no GPU required. The first run downloads the model; subsequent runs read it from the cache, so the second invocation completes in ~1 second on a 1024×1024 input.
No. This preset runs entirely in your browser using WebAssembly + canvas APIs — no server round-trip. You can verify this in your browser's network panel: only static asset requests, no image upload. The file never leaves your device.