Local Cutout — Image Background Remover

Remove backgrounds locally, refine with brushes, and save transparent PNGs. No uploads, accounts, or API keys.

Local Cutout — Image Background Remover Remove backgrounds from your images locally in Chrome, then save transparent PNGs. Local Cutout includes its subject-detection model inside the extension. Your photos are processed on your device: no account, API key, image upload, or processing server is needed. WHAT YOU CAN DO • Use Smart subject removal for photos, products, people, and characters. • Use Color removal for simple graphics and plain backgrounds. • Refine the result with Erase and Restore brushes. • Undo up to five edits and compare with the original. • Preview transparency against checkerboard, white, or dark backgrounds. • Export a transparent PNG. HOW TO USE IT 1. Click the extension icon to open the editor. 2. Choose or drop a PNG, JPEG, or WebP image. 3. Leave Smart subject removal selected and click Remove background. 4. Refine any imperfect areas, then select Save transparent PNG. PRIVACY The extension uses only images you explicitly select. Images and edit masks are processed locally. The extension does not upload photos, collect analytics, track browsing, or require a login. Clear, close, or reload the editor to discard the editing session. Your downloaded PNG stays on your computer. PLEASE KNOW BEFORE INSTALLING • The download is approximately 168 MB because the complete local model and runtime are bundled. • Smart removal uses CPU and memory. Processing can take tens of seconds or longer on slower computers. • Images are limited to 20 MB and resized to a maximum 2,000-pixel longest side. Exports use these working dimensions. • The subject mask is predicted at 512 × 512. Hair, fur, glass, tiny details, and ambiguous subjects can need manual cleanup. • Keep only the main subject is enabled by default. Turn it off to retain several separate predicted subjects. • There is no autosave, batch mode, or saved-project feature in this version. The extension includes the MIT-licensed BiRefNet-lite 512 model and ONNX Runtime Web. License notices and attribution are included in the package.