What does this background remover do?

This tool strips backgrounds from images without sending anything to a server. Pick a color mode to delete a solid backdrop, grab the magic wand to cut out a region by color similarity, click an object to select it precisely, or switch on the Smart AI mode to matte the subject automatically. You can then export a transparent PNG, or drop in a solid color or a photo as the new background, which is exactly what ID photos and product shots need.

Key features

Pick object: click something in the picture and a Segment Anything model running on your device traces its outline, textures and all. Shift+click (or right-click) marks a spot that is not part of the object when the selection grows too far.
Cut out or keep: every selection can either be removed or be the only thing kept, so isolating one subject out of a crowded photo is a single toggle.
Solid color removal: delete any single background color you pick.
Magic wand: click to select a color region; contiguous (connected) or global (all matching colors); click again to add more regions.
Smart AI matting powered by an IS-Net ONNX model running locally via ONNX Runtime Web.
Edge cut and feather controls to refine the AI matte.
Background replacement: transparent, solid color presets (white, black, blue, red for ID photos), or a custom uploaded image.
Export to PNG, JPEG, or WebP, with adjustable quality.
Fully local: the image is decoded and processed in your browser tab.

How to use it

Drop an image onto the upload area or click to browse.
Choose a mode: Solid color, Magic wand, Smart AI, or Pick object.
- For Solid color, pick the color and tune the tolerance.
- For Magic wand, click the preview to cut a region; raise the tolerance or disable "contiguous only" to widen the selection.
- For Smart AI, click "Remove with AI"; tweak Edge cut and Feather if needed.
- For Pick object, wait for the one-time model load, click the object, then add correction clicks until the outline is right. "Add another object" locks in the current one so you can pick the next.
For the wand and Pick object modes, choose whether the selection is cut out or is the only thing kept.
Pick a background: transparent, a solid color (try the ID-photo presets), or an uploaded image.
Choose the export format and quality, then download the result.

Removing versus keeping an object

The two directions solve opposite problems. "Cut out the selection" deletes what you picked and leaves the rest of the photo alone, which is how you delete a passer-by, a logo, or a distracting prop. "Keep only the selection" throws away everything else, which is how you lift one product, one pet, or one person out of a busy scene onto a transparent background. Both work on the same selection, so you can flip between them without re-clicking.

Which format should you pick?

PNG: lossless with real transparency, best for cutouts, logos, and further editing.
WebP: smaller than PNG and supports transparency.
JPEG: smallest for photos, but transparency is flattened, so a background color is baked in.

Why remove a background?

Transparent cutouts let you place a subject on any backdrop, build consistent product catalogs, or produce correctly backed ID photos (white, blue, or red). The magic wand handles banners and flat-color scenes instantly, AI matting tackles hair, fur, and busy backgrounds that color tools cannot, and object picking covers the case where the thing you want is not the background at all.

Privacy and security

Every pixel is decoded, cut, and re-encoded in your browser using the Canvas API and ONNX Runtime Web. Nothing is uploaded to any server. Object URLs are revoked when you clear the image or leave the page, and models are cached locally after their first download.

FAQ

Does the AI mode need a server?

No. The matting model and the object-picking models all run in your browser through ONNX Runtime Web. Each is downloaded once from a public CDN and then reused from cache, and your image is never part of any request.

How large are the model downloads?

Smart AI matting needs about 42 MB, an int8-quantized IS-Net. Pick object needs about 13 MB in total: an 8.5 MB image encoder plus a 4.7 MB mask decoder, both int8-quantized SlimSAM. You only pay for the mode you use, each model downloads once, and the browser caches it afterwards. For Pick object the encoder runs once per image and the decoder runs per click, so refining a selection is fast after the first wait.

Why does the outline sometimes include too much?

One click is a hint, not an instruction. Shift+click (or right-click) on the part that should not be included and the outline is recomputed with that exclusion. Two or three clicks resolve most cases.

Can I keep transparency for a product photo?

Yes. Leave the background on "Transparent" and export PNG or WebP. JPEG will flatten the background into a solid color.

Is there a size limit?

Processing is local, so the practical limit is your device memory. Typical phone and camera photos are handled fine. Both AI modes resize to a fixed 1024x1024 before inference, so the model itself costs the same at any resolution; what grows with a very large image is the decoding, the full-resolution matte, and the re-encode.

Image Background Remover - Objects, Wand, AI