Is the image stored?
No. The image is processed in memory by Tool Teaze's isolated inference worker, streamed back and discarded. Tool Teaze does not retain it or use it for model training. Manual edits remain in your browser.
Create an automatic foreground cutout with Tool Teaze's isolated inference worker, then erase or restore edges, replace the background, adjust the subject and export the result.
JPG, PNG or WebP · up to 22 MB and 50 megapixels
Choose an image, paste one, drop one here, or enter a public HTTPS image URL.
Privacy: the selected image is sent only to Tool Teaze’s isolated inference worker, processed in memory, streamed back, and not retained or used for training. Manual edits stay in this browser. URL imports are restricted to public HTTPS images.
The pinned MIT-licensed BiRefNet Lite model predicts a foreground mask. Fine hair, glass, reflections, shadows, crowded scenes and low-contrast edges can need manual correction. Review every result.
The worker accepts one still JPG, PNG or WebP up to 22 MB, 50 megapixels and 25,000 pixels per edge. It uses a bounded queue and one inference at a time. Cancelling closes the browser request; already-started inference may finish inside the worker and is immediately discarded.
Automatic removal runs on Tool Teaze infrastructure because the measured browser-only candidates exceeded the authorised 512 MiB web-memory ceiling. The main web container keeps its existing limit.
Background Remover sends one validated image to an isolated Tool Teaze worker running a pinned BiRefNet Lite model, streams back a transparent PNG, and provides local cutout, background, effects, adjustment and design controls.
Input: A product photo on a cluttered table
Output: A reviewed product cutout on transparency or a replacement background
No. The image is processed in memory by Tool Teaze's isolated inference worker, streamed back and discarded. Tool Teaze does not retain it or use it for model training. Manual edits remain in your browser.
Measured browser-only BiRefNet and ISNet candidates exceeded the authorised 512 MiB web-memory ceiling. Isolating inference in a separately bounded worker keeps the main site limit unchanged and avoids pushing a large model into every visitor's device.
Yes. Use Erase to remove more foreground and Restore to bring source pixels back. You can undo and redo recent brush strokes.
Keep transparency, choose white, black or any colour, blur the source background, or upload a replacement JPG, PNG or WebP image.
Download the original-resolution transparent PNG returned by the worker, or export the edited design as PNG. JPEG is enabled when you choose an opaque background.
No. Fine hair, glass, reflections, shadows, crowded scenes and low-contrast edges may need manual correction. Always review the preview.
The worker accepts one still JPG, PNG or WebP up to 22 MB, 50 megapixels and 25,000 pixels per edge. A bounded queue protects service availability.
The worker uses a pinned FP16 ONNX conversion of BiRefNet Lite under MIT. The model identity, SHA-256 hash, licence and source notices are included with the service.