AI Background Remover
Create a transparent PNG using a locally running Apache-2.0 background-removal model. The browser downloads a model from Hugging Face and runs it on this device; your input is not sent to ToolkitPoint.
How to use AI Background Remover
- Choose a PNG, JPEG or WebP photo under "Choose an image"; a plain, contrasting background works best.
- Press "Remove background" and wait for the ORMBG model to download; a progress bar tracks the first load.
- Watch the status text: it changes to "Separating foreground from background…" while the model runs on your device.
- Use "Stop waiting" if it takes too long or "Reset" to choose a different picture.
- Review the result on the checkerboard preview, zoom into hair and edge areas, then press "Download transparent PNG".
- For full-resolution output, brush touch-ups, replacement backgrounds and batches, use the section below the result: choose photos, press "Remove backgrounds (full resolution)", fix edges with the Erase and Restore brush, pick a background and download PNG, WebP or JPEG (or a ZIP).
Worked example
Drop a 3000 x 2000 product photo into the full-resolution section. The result is 3000 x 2000 with a transparent background; brush away a leftover patch, choose a white background and crop to the subject, then download a JPEG (or the whole batch as a ZIP).
When to use this tool
- Cutting out a product for an online shop listing.
- Isolating a person from a busy background for a profile image.
- Making a sticker-style cutout for a presentation slide.
Important limitations
The top button works on a 1024 px copy; the full-resolution section below it enlarges the mask to the original size. The ORMBG model requires a substantial first download and can be slow or fail on memory-limited devices. Fine hair, glass, shadows, motion blur, and low-contrast edges may be cut incorrectly.
Frequently asked questions
Does it keep my full resolution?
The original button above still works on a copy reduced to 1024 pixels. The full-resolution section enlarges the AI mask to your photo's real size, snaps its edges to the full-size picture with a guided filter, and applies it to your original pixels, so a 4000 x 3000 photo comes back 4000 x 3000 (up to 24 megapixels; edge refinement is done up to 12 megapixels).
What kind of pictures does it struggle with?
Fine hair, glass, smoke, motion blur, shadows and low-contrast edges may be cut too much or too little. Always zoom in on the edges; a manual touch-up may still be needed.
Why does the first run take so long?
The model file is fairly large and downloads on first use from a public model host. On phones with limited memory, it can fail or stall, and that is why there is a "Stop waiting" button.
What if it finds no background to remove?
After running, the tool inspects the mask, and if almost nothing is transparent it shows an error instead of a file. Try a subject with clearer separation, then download background-removed.png when it succeeds.
Can I fix mistakes and change the background?
Yes. In the full-resolution section, Erase and Restore brushes (with size and softness) correct the mask, and Edge feather, Edge shift and the colour-halo option clean the outline. You can save with a transparent, solid, gradient, blurred-original or your own image background, add a soft shadow, and crop to the subject.
Is the edge refinement an AI matting model?
No. It is a standard guided filter that snaps the enlarged mask to edges in your full-size photo, and the model runs in WebAssembly (not on the GPU). It cannot recover fine hair the model missed, so review edges and use the brush.