Click to upload or drag & drop images
JPG, PNG or WebP — select multiple files at once
Two modes: color keying and AI subject detection
Color key samples the color at the top-left corner of your image (or one you pick) and makes every pixel within your tolerance of that color transparent. It's instant and gives pixel-perfect edges on logos and product shots against a plain backdrop. AI subject detect runs the open-source U²-Net segmentation model directly in your browser via WebAssembly: it detects the subject's outline regardless of background color, which is what you need for photos with busy, gradient, or multi-colored backgrounds. Both modes process the image entirely on your device — nothing is uploaded to any server.
Which mode should you use?
If the background is a single flat color — a studio backdrop, a white product sweep, a logo canvas — use color keying: it's instant and its edges are exact. If the background is a room, outdoors, or anything with multiple colors, use AI mode: Fast and High quality run at 320px, while Best 1024² sees ten times the detail. Whichever model you pick, the resulting mask is sharpened back up against your original photo using an edge-aware filter — not a plain blur-and-stretch — so edges follow real detail like hair and fabric instead of smearing. Heavyweight commercial services still run larger models on servers, which is the trade-off this tool deliberately avoids by keeping your photo on your device. Try a couple of modes on the same image and keep whichever result is cleaner. If the AI still misses a spot, switch on the Erase or Restore brush in the preview and paint over it by hand before downloading — no model gets every edge perfect on its own.
Frequently asked questions
How does automatic background removal work here?
Two ways, your choice. Color key mode samples a background color and makes every pixel within your chosen tolerance transparent — best for plain backdrops. AI mode runs the U²-Net segmentation model in your browser via WebAssembly to detect the subject's outline — best for busy photo backgrounds.
Why didn't it remove the whole background, or why did it remove part of my subject?
In color mode, click directly on any leftover background area in the preview — each click selects that color for removal, so a background with several distinct colors can be keyed out with a few clicks. If it's eating into your subject instead, lower the tolerance slider. In AI mode, try the High quality model and the edge-sharpness slider, or switch on the Erase/Restore brush to correct the exact spot by hand.
Can I fix mistakes in the AI result by hand?
Yes — in AI mode, the preview has an Erase/Restore brush under the edge-sharpness slider. Erase paints leftover background back to transparent; Restore brings back subject detail the model missed (stray hair, a faint edge). Drag over the preview to paint, adjust brush size as needed, and Reset touch-ups if you want to start over — your hand-painted areas are kept even if you then change the AI model or the sharpness slider.
Will this work on photos with busy or gradient backgrounds?
Yes — switch to AI subject detect mode. Color keying only suits solid backgrounds, but the AI model detects the subject's outline regardless of what's behind it. As a compact on-device model it can miss very fine edge detail (stray hairs, fur wisps), so check the preview before downloading.
Is the AI processing done on a server?
No. The first time you use AI mode, your browser downloads the runtime and your chosen model — roughly 18 MB for Fast, 57 MB for High quality, 115 MB for Best 1024² — and caches them. Detection then runs on your own device (on your GPU via WebGPU where supported) — your photo never leaves it, which you can verify from the Network tab in DevTools.
Can I get a solid white or colored background instead of transparent?
Yes — in the preview, the "Output background" option under the accept button lets you choose Transparent (default), White, or a custom color. Passport/ID photos, marketplace listings (Amazon requires a pure white background for its main product image), and similar use cases need a genuinely solid, non-transparent background — this flattens the result onto that color rather than leaving it as alpha transparency.
What image formats does this work best for?
The result downloads as a PNG — transparent by default, or with a solid background if you choose one. It works from any JPG, PNG, or WebP source image.