AI Background Remover & Cutout Studio
Remove image backgrounds locally in your browser using open-source segmentation models (BiRefNet Lite MIT / MODNet Apache-2.0) or instant studio keying. Export full-resolution transparent PNGs or composite onto white, solid-color, or custom image backdrops.
Method and assumptions
Images are decoded at their native pixel dimensions in browser memory. When BiRefNet Lite (onnx-community/BiRefNet_lite-ONNX, MIT License) or MODNet (Xenova/modnet, Apache-2.0 License) is selected, @huggingface/transformers runs the ONNX segmentation graph on WebGPU (when supported by your GPU/browser) or falls back to WebAssembly, caching model weights in your browser Cache API. The predicted alpha matte is upsampled to the original image dimensions and applied to the full-resolution RGB buffer before compositing onto a transparent, white, solid-color, or custom image backdrop.
Common mistakes and edge cases
BiRefNet Lite downloads ~109 MB of FP16 ONNX weights on first use (cached thereafter); MODNet downloads ~6.3 MB for faster CPU/mobile execution; Studio Backdrop Keyer runs instantaneously (0 MB download) for uniform studio backdrops. Semi-transparent glass, fine hair against low-contrast backgrounds, and heavy motion blur may leave soft halo edges.
How to use
- Drop one or more JPG, PNG, or WebP images into the workspace (or load the sample studio product).
- Choose your segmentation engine (BiRefNet Lite for general/product high-res cutouts, MODNet for fast portrait matting, or Studio Backdrop Keyer for instant uniform-backdrop removal).
- Select Transparent PNG, Pure White (#FFFFFF), Custom Solid Color, or upload a replacement background image.
- Inspect the before/after slider and download individual PNGs or a batch ZIP archive.
Example
Load the sample studio watch product or drop a portrait/product photo, select Transparent PNG or Pure White (#FFFFFF), inspect the before/after slider, and export the full-resolution cutout PNG or batch ZIP.
Frequently asked questions
- Are my photos sent to an external AI API?
- No. Model weights are downloaded to your browser's Cache API and executed locally on your device using WebGPU or WebAssembly.
- Does this preserve my original image resolution?
- Yes. The predicted alpha matte is upsampled to match your source image's exact width and height and applied directly to the original full-resolution pixels.
RELATED TOOLS
- PDF & Image OCR Text Extractor
Extract text from scanned PDFs and images locally with multi-language Tesseract OCR.
FROM THE PORTFOLIO
Explore the RoninDesk portfolio.
This tool is provided for general informational and workflow purposes only. It does not constitute financial, investment, tax, or professional advice. All calculations run in your browser; no input data is transmitted or stored.