Background Remover
AI-powered background removal that runs entirely in your browser. No upload, no watermark, full resolution.
100% in your browser — files never leave your device
How on-device background removal works
Most background removal services send your photos to a remote API, charge per image, and retain rights to use uploads for model training. This tool takes the opposite approach: the entire inference pipeline runs inside your browser.
When you drop an image, the tool checks whether the selected model is already in your browser cache. If not, it downloads the quantized ONNX model (~42-43 MB) once and stores it using the Cache API. All subsequent uses load from local storage — no network request required.
Inference runs via the ONNX Runtime Web backend. On devices with WebGPU support, the model executes on your GPU for fast results. On older hardware, the CPU backend handles it automatically. Both models process images at 1024x1024 resolution. The "People" model (RMBG-1.4) is tuned for portrait hair and skin edges; the "General" model (IS-Net) handles objects, product photos, and graphics more broadly.
After removal, you can choose a background color, fine-tune the mask edge with the refinement controls, or switch models and reprocess — all without re-uploading your image.
Frequently asked questions
Are my images uploaded to a server?
No. The AI model runs entirely inside your browser using WebGPU acceleration (with automatic CPU fallback). Your images never leave your device — no upload, no account, no size limit.
Which AI model is used for background removal?
Two models are available. "People" uses BRIA RMBG-1.4 (Apache 2.0, ~42 MB quantized ONNX) — optimized for portraits and hair detail. "General" uses IS-Net (Apache 2.0, ~43 MB quantized ONNX) — better for objects, products, and graphics. Both run at 1024x1024 input resolution.
Why does the first run take a moment?
The model file (~42-43 MB) downloads once from a CDN on first use and is then cached in your browser via the Cache API. Subsequent uses are instant — the model loads from the local cache without any network request.
What is the difference between WebGPU and CPU inference?
WebGPU runs the neural network on your GPU, which is typically 5-15x faster than CPU inference for ONNX models at 1024x1024. The backend used for each image is shown in the result view. CPU fallback is automatic when WebGPU is unavailable.
Can I adjust the mask after processing?
Yes. The mask refinement controls let you tune edge smoothing and threshold after the initial result. You can also switch models and click "Re-process" to run the other AI model on the same image without re-uploading.
Related tools
Image Compression
Compress to WebP, AVIF, JPEG, PNG, or JPEG XL with full quality control.
Color Correction
Brightness, contrast, exposure, temperature, and more. Real-time WebGL2.
Convert PNG to WebP
Convert PNG to WebP for smaller file sizes
EXIF Viewer & Remover
See camera, timestamp, and GPS data — then strip it before sharing.