AI Background Removal
Online AI background removal tool: powered by deep-learning segmentation models (RMBG / MODNet), it automatically detects and cuts out people, game character art, icons and product subjects right in your browser — even from photos, gradients or textured backgrounds — and exports transparent PNGs. Two models are provided: General for objects, icons and game assets, and Portrait optimized for people/characters. WebGPU acceleration is used when available, with automatic WASM fallback. Supports batch processing, manual eraser touch-up and ZIP download. Models and images run entirely locally in your browser, never uploaded.
Models download only once on first use, then load instantly from your browser cache — no more waiting on subsequent runs.
Pick by subject type: General (RMBG) suits objects, icons and game assets; Portrait (MODNet) is optimized for people and characters. WebGPU is used when available, with automatic WASM fallback. Everything runs locally.
Features
- One-click AI cutout: auto-detects the subject, no color picking, works on complex/photo backgrounds
- Two model tiers: Fast (~44MB, general subjects) and Fine (MODNet, ~26MB, portrait/character optimized)
- WebGPU acceleration first, automatic WASM fallback for wider compatibility
- Runs locally: models and images are inferred in your browser, never uploaded
- Eraser for manual touch-up, live transparent / solid-color preview
- Batch process multiple images and download transparent PNGs as a ZIP
FAQ
Are my images uploaded to a server?
No. The AI model runs inference locally in your browser. Images are never uploaded, so the whole process stays private and secure.
Why do I have to wait for a download on first use?
AI matting needs to download a segmentation model (Fast ~44MB, Fine ~176MB). It downloads once on first use and is cached by the browser, so subsequent runs are much faster.
General vs Portrait — which should I pick?
General is based on RMBG, great for icons, products, game assets and batch work. Portrait is based on MODNet, optimized for people/characters with more natural edges. Both are compact and prefer WebGPU acceleration.
How is this different from the Chroma-Key / Dual-Background tool?
Chroma-key / dual-background matting relies on solid or green-screen backgrounds and cannot handle complex ones. AI matting uses deep learning to understand subject vs. background, so it works on photos, gradients and textures without requiring a solid background.
Why is it slow in my browser?
When the browser does not support WebGPU it falls back to WASM, which is slower. Use a recent Chrome / Edge to enable WebGPU acceleration.