How to remove the background
- 01
Choose a clear portrait
Use even lighting and a subject that is visibly separated from the background.
- 02
Run local removal
The model loads on demand in a worker and creates an editable mask preview.
- 03
Refine and export
Adjust edge shift, feather, opacity, and background conservatively before downloading.
Best inputs and exports
- JPG, PNG, or WebP portraits up to the published file and pixel limits
- Transparent, white, black, custom-color, or blurred-original backgrounds
- PNG, WebP, or JPG full-size export
Model limits
- The model is trained for people and portraits, not general products, animals, or every foreground object.
- Hair, veils, transparent objects, motion blur, and low-contrast edges may need careful refinement.
- JPG cannot preserve transparency and is exported with an opaque background.
Check difficult edges
Inspect hair, shoulders, fingers, and any region where foreground and background have similar colors at a large preview size.
- Compare both light and dark backgrounds.
- Avoid large edge adjustments that erase hair.
- Open the exported file to confirm transparency or the chosen color.
If the cutout is weak
- Try a better-lit portrait with more contrast around the subject.
- Reduce edge adjustments and add only a small amount of feathering.
- Use a smaller image if the browser reports a memory limit.
Questions about portrait removal
Does the portrait leave this device?
No. The source pixels, model inference, mask, compositing, and export remain in the browser. Model and runtime downloads are separate requests.
Will it work on products or pets?
It may produce a result, but this model is scoped to people. A product, animal, or complex scene is outside the promised use case.