How to remove an image background locally

Removing a background is useful when you want a clean product image, a cutout for a presentation, or a photo of a pet without the clutter. TinyModel processes the image in your browser. The model downloads to your device; your photo is not sent to a processing service.

Start with a suitable image

Choose a JPG, PNG, or WebP with a clearly visible subject. Strong contrast between the subject and background usually makes the boundary easier to identify. Fine hair, transparent glass, motion blur, and objects partly hidden behind other objects are more demanding.

Keep your original. A cutout is an interpretation of which pixels belong to the foreground, and some edge detail can be lost. If the image has text or a logo, inspect those details before using the output.

Make your first cutout

Open Background Remover.
Choose an image. The workspace shows a local preview and the model download information.
Select Remove background. The first run includes loading the model and preparing your GPU.
Use the Original and Result buttons to compare the whole image.
Download the transparent PNG when you are happy with the result.

The checkerboard behind the result indicates transparency. It is part of the preview, not part of the exported image.

Understand the first download

A local model must be available in your browser before it can process anything. BiRefNet Lite 512 uses a reduced input resolution to make browser inference practical. The model weights are around 94 MB, with additional runtime assets. Those files are application assets, not copies of your image.

Your browser can cache these assets. Clearing site data or model caches means they need to be downloaded again. Cached models are not a promise that the entire website works offline after a restart.

Inspect the edges

Check hair, fur, gaps between fingers, straps, plant stems, and translucent objects. The model predicts a mask at its inference resolution and maps it back onto the original image. Large exports retain the original dimensions, but that does not restore detail missing from the predicted mask.

For difficult scenes, try a closer crop of the subject before processing. If the cutout is not good enough, retain the original for a tool that offers manual mask editing. TinyModel does not currently provide a manual refinement brush.

If processing does not start

This tool requires WebGPU and hardware acceleration. A recent desktop browser is the best starting point. Device memory and GPU capabilities also matter. If initialization fails, close other demanding tabs and try a smaller image or another supported browser. The workspace never switches to uploading the image as a fallback.

Before sharing

Review the PNG against both light and dark backgrounds in your destination app. Confirm you have the rights to use the original image, and keep an untouched copy for future edits.