Remove an image background privately

Remove distracting backgrounds while keeping the foreground subjects on your device. FileGizmo runs an image-segmentation model locally and exports transparent PNGs. Choose one image for manual refinement or queue up to 20 images for sequential processing and one ZIP download. Clear edges, good lighting, and visible contrast between subject and background improve the mask.

Quality mode uses BiRefNet Lite, an MIT-licensed segmentation model packaged with FileGizmo for local browser inference. The first run loads a large model from the FileGizmo site into browser storage, which can take noticeably longer than a conventional upload tool. The photograph itself is not sent with that request. Later visits can reuse cached model assets until the browser clears them. Fast mode uses a smaller portrait-focused model when download size and turnaround matter more than the hardest edges.

Batch mode accepts up to 20 images and processes them sequentially to control memory use. Manual refinement remains a single-image workflow because each mask needs visual review.

Background removal predicts which pixels belong to the foreground; it does not understand the scene with human certainty. Strong contrast, even lighting, and a complete visible subject help. Fine flyaway hair, transparent glass, motion blur, smoke, fur, shadows, and foreground colors that closely match the background are difficult cases. The real-photo QA evidence includes both loose hair and a translucent product rather than relying only on simple synthetic shapes.

Inspect the checkerboard preview around hair, fingers, product handles, and gaps inside the subject. A transparent result can look clean on white while revealing a halo on dark colors, so test it against the background where it will actually be used. Choose Refine edges when the automatic mask misses: Erase removes leftover background, while Restore brings original pixels back. Both tools edit the alpha mask rather than damaging the source image.

The brush has a soft edge and a 4–80 pixel size control. Use the wheel or a two-finger pinch to zoom, switch to Pan to move around the image, and hold Compare to see the original. Undo and redo keep the latest 20 strokes. FileGizmo retains the original pixels and automatic mask in browser memory while you edit, then re-encodes the full-resolution PNG only when you download.

The output is PNG because JPG cannot store transparency. PNG may be larger than the input even though the background is gone; background removal and file compression solve different problems. Resize the cutout to its real display dimensions and then use an appropriate PNG or WebP workflow if delivery size matters. Keep the original photo as a separate source even though Restore can recover pixels during the current editing session.

Inference happens in a worker so the page remains responsive, but machine-learning models still use substantial memory and CPU. A desktop with a current browser provides the most reliable experience for large photos. On mobile, close other tabs and avoid switching away during processing. Reduced speed is a local-device cost, not an upload queue.

The model files are third-party software assets, not a hidden photo-processing API. Quality mode is packaged from the MIT-licensed BiRefNet Lite model and loaded only on the background-removal page after user intent. FileGizmo’s license inventory records the model source and local deployment. The browser can cache those assets, but private browsing modes and storage cleanup may require another first-load download.

Real-photo QA uses a person with loose hair and a translucent bottle among similarly colored objects. The fixtures confirm a working alpha output while documenting that faint flyaways and ambiguous edges may still need a Restore or Erase pass.

Use background removal as a starting point for a product listing, profile image, slide, thumbnail, or design composite. It is not evidence that an object was photographed on transparency, and it should not be used to misrepresent products or remove legally required context. Review the entire cutout, not just the most obvious central subject.