Koutuxia removes the background from a photo and hands you a transparent PNG. The free path runs BiRefNet-lite in the browser: a 47 MB int8 model downloaded once from a CDN and cached, then executed through WebGPU where the device supports it and WASM SIMD where it does not, at a smaller input resolution on weaker hardware. A paid path sends the image to a cloud model instead, for finer edges, higher resolution and batch jobs. Around the cutout you also get a one-click background colour swap for red, blue and white, a photo restoration mode, and direct export to transparent PNG. The browser-side tools are free and need no signup, and the site works from mainland China without a proxy.
BiRefNet-lite cutout that runs in the browser, so the image never leaves the machine on the free path
47 MB int8 model downloaded once from a CDN and cached for later visits
WebGPU where the device supports it, WASM SIMD where it does not, at a smaller input resolution on weaker hardware
Paid cloud path for finer edges, higher resolution and batch jobs
One-click background colour swap for red, blue and white
Photo restoration mode for blurry or low-resolution images
Direct export to transparent PNG
Browser-side tools are free and need no signup
Works from mainland China without a proxy
Getting a transparent PNG of a logo or product shot without uploading the original anywhere
Swapping an ID or profile photo background to red, blue or white
Cleaning up a product photo before it goes on a store page
Sharpening an old or low-resolution photo before cutting it out

The default for background removal is to upload your photo to someone else's server. The free path here does not: BiRefNet-lite runs in the browser through WebGPU, or WASM SIMD when the device has no WebGPU, so the image stays on the machine. The 47 MB model downloads once and is cached. There is a cloud path for finer edges and batch work, but the browser-side tools stay free and need no signup. Curious which images break the local model for you.

The default for background removal is to upload your photo to someone else's server. The free path here does not: BiRefNet-lite runs in the browser through WebGPU, or WASM SIMD when the device has no WebGPU, so the image stays on the machine. The 47 MB model downloads once and is cached. There is a cloud path for finer edges and batch work, but the browser-side tools stay free and need no signup. Curious which images break the local model for you.
Find your next favorite product or submit your own. Made by @FalakDigital.
Copyright ©2025. All Rights Reserved