← Back to Home

Background Remover

Remove an Image Background

Select a JPG, PNG or WebP image. The AI model runs in your browser and creates a transparent foreground cutout.

Choose an image or drop it here JPG, PNG or WebP · up to 25 MB · image limits adapt to your device
No image selected.
0%
Increase only when faint background pixels remain around the subject.
Trim empty background around the detected foreground while keeping the subject at full available resolution.

First use downloads the browser AI runtime and an approximately 44 MB quantized model. They can be cached by your browser for later use.

About

XAVERT Background Remover isolates the main foreground subject of an image with browser-based AI and creates a cutout that can be exported with transparency or a solid background.

Main Features

  • Automatic AI background removal for JPG, PNG and WebP images.
  • Full-resolution output within browser-safe image limits.
  • Transparent, white or custom-color backgrounds.
  • Optional edge cleanup for faint background remnants.
  • Automatic subject cropping with adjustable output padding.
  • PNG, WebP and JPG downloads with configurable export quality.
  • Before-and-after preview and PNG clipboard copy when supported.

Privacy First

Your selected image is processed in your browser and is not uploaded to XAVERT. On first use, your browser downloads the AI runtime from jsDelivr and the ORMBG model files from Hugging Face. Those external resources may be cached by your browser.

Runtime: Transformers.js 4.2.0 (Apache-2.0). Model: onnx-community/ormbg-ONNX (Apache-2.0). XAVERT currently uses the quantized q8 model in WASM compatibility mode because background-removal WebGPU support can vary across browser and driver combinations.

Important Note

AI cutouts can require manual refinement around hair, glass, shadows, transparent objects or very similar foreground and background colors. Edge Cleanup can make uncertain edges crisper, but high values can also remove fine details. Auto Crop uses the resulting alpha mask and may leave extra space when the model detects soft shadows or semi-transparent edges.