- Are my images uploaded to a server?
- No. The background removal runs entirely inside your browser using on-device AI. Your photo is never sent anywhere — not to us, not to anyone. The only network traffic is the one-time download of the AI model itself (see below).
- What is the AI model, and why does it need to download?
- The tool uses BiRefNet-lite, an open-source (MIT-licensed) image segmentation model. On your first use, it is downloaded once from Hugging Face's CDN (roughly 100–190 MB depending on your hardware) and then cached by your browser, so later visits start instantly. Importantly, only the model travels to you — your image data never travels to anyone.
- What is the difference between Fast and Best quality modes?
- Fast uses the BiRefNet-lite model (~95 MB download) and returns results in seconds — ideal for everyday photos. Best switches to the full BiRefNet architecture (~950 MB download, run on your CPU for maximum accuracy and consistency), which resolves finer details on complex scenes like patterned clothing, hair, or foliage, but takes noticeably longer per image. Both run entirely on your device and are cached after the first download; your choice is remembered for next time.
- The AI missed a spot or removed too much — can I fix it?
- Yes. After the background is removed, click "Touch Up Manually" to paint directly on the result: the Erase brush removes leftover background fragments, and the Restore brush brings back subject areas that were incorrectly removed. You get an adjustable brush size, undo history, and the download always reflects your edits. Even the best automatic tools struggle with color-similar subjects and busy backgrounds — that's exactly what the touch-up brush is for.
- Which image formats are supported?
- You can import JPG, PNG, and WebP images. Output is always a PNG with a transparent background. HEIC/HEIF photos (common on iPhones) are not supported yet — convert them to JPG first.
- Is there a file size or resolution limit?
- Files up to about 30 MB work best. Images larger than 4096 pixels on their longest edge are scaled down automatically before export so the browser stays stable — you'll see a notice when this happens.
- How good is the quality?
- The model excels at clear subjects: people, products, animals, and objects with well-defined edges. Busy scenes with hair, fur, or semi-transparent details may need touch-ups in an editor. The before/after slider lets you check the result before downloading.
- Which browsers work best?
- Chrome and Edge on desktop offer GPU-accelerated processing via WebGPU (fastest). Firefox and Safari fall back to CPU mode, which produces the same quality but takes longer per image.
- Does it work offline?
- After the first visit, the model is cached by your browser, so repeat runs are fast and mostly offline-friendly — though the page itself still needs a normal page load.
- Is it free? Can I use the results commercially?
- Yes. The tool is free with no sign-up and no watermarks, and because the model is MIT-licensed, you can use your exported images however you like — including commercially.