Remove image backgrounds with AI — 100% in your browser
High detail uses BiRefNet-lite (MIT) on WebGPU — best for hair, fur and product edges. Downloads ~115MB once, then cached.
Upload an image and click "Remove Background"
Then choose a new background to replace it
Background removal runs two permissively licensed segmentation models in your browser. BiRefNet-lite (MIT, ~115MB) needs WebGPU and handles hair, fur and products; MODNet (Apache-2.0, ~7MB) is the CPU fallback for portraits. Weights are downloaded once and cached. No data leaves your device.
Background removal is semantic segmentation: a segmentation network (BiRefNet for photos, MODNet for portraits) is trained to label every pixel as "foreground subject" or "background". The model runs in your browser through ONNX compiled to WebAssembly — the image is decoded, passed through the network, and the network outputs a soft alpha matte (a 0–255 opacity layer) that keeps fine edges like hair. That matte turns your photo into a transparent PNG you can drop onto any new background.
The pipeline: the segmentation network isolates the subject, produces an alpha matte that preserves soft edges, and the foreground is composited onto a transparent, solid-color, gradient, blur, or custom background.
Nina photographs a handbag on a busy office desk for her e-commerce store. She uploads the photo, keeps the default High detail model, and BiRefNet segments the bag while keeping the soft shadow and strap edges intact.
To remove a background: upload your image, the AI model runs locally in your browser and segments the subject from the background, producing a transparent PNG — no data uploaded.
FreeToolHub AI Background Remover is a free browser-based tool that removes image backgrounds using AI, no signup, no upload.
Remove image backgrounds with AI — 100% in your browser, no upload, no server. Replace with transparent or custom background. Free.
The AI Background Remover is a free tool that erases the background from photos without uploading a single pixel. It runs a neural network — BiRefNet-lite or MODNet, executed through ONNX Runtime directly in your browser — so the image is processed on your own hardware and never touches a server. High detail (BiRefNet-lite, MIT, about 115MB) preserves fine hair and product edges on WebGPU; Portrait (MODNet, Apache-2.0, about 7MB) is a fast CPU fallback for people. Beyond plain transparency, the tool can composite your cut-out onto solid colors, preset gradients, a blurred version of the original, or a custom image you upload, and exports a ready-to-use transparent PNG.
E-commerce sellers use it to meet marketplace rules that demand white or transparent product backgrounds, cleaning up dozens of listings without paying per image. Job seekers and professionals cut themselves out of casual photos to make presentable headshots for LinkedIn. Graphic designers isolate subjects for thumbnails, posters, stickers, and composite artwork, using the gradient and custom-background modes for instant mockups. Photographers and agencies bound by NDAs appreciate that client portraits never leave the machine. Small business owners produce consistent product catalogs, while students and teachers lift subjects out of photos for presentations and collages. Anyone who has waited in an upload queue for a simple cut-out will notice the difference immediately.
(1) Drop in a JPEG, PNG, or WebP image — the file is checked for size limits before anything starts. (2) Pick a model: Standard, which is fast and lightweight, or HD, which keeps hair and edges sharper. On first use the model downloads with a live progress bar, then stays cached for every later session. (3) Click Remove Background. The segmentation runs on your device, and a before/after slider lets you drag across the result to inspect the mask. (4) Choose a replacement: keep it transparent, fill with one of nine preset colors, apply a gradient, blur the original scene by an adjustable amount, or upload your own backdrop, then download the finished PNG.
Use High detail for product shots and anything with fine edges — strands of hair, fur, lace, grass, glassware rims, or thin branches. Use Portrait when you are on a device without WebGPU or want a 7MB download: MODNet is tuned for people and loads in seconds, but it is not intended for products. Both models run locally and are free with no per-image limits. Two more tips: images with strong contrast between subject and background separate cleanly, while low-light phone photos with soft edges may need a second pass. If a download stalls, the tool warns you after 90 seconds and suggests retrying or switching models. Both models are free with no per-image limits.
Genuinely strong - the BiRefNet-lite model this tool runs was trained precisely on matting boundaries like hair, fur, and semi-transparent edges, and it produces soft alpha at the edge instead of a hard cutout. Results degrade on low-contrast subjects (white shirt on white wall) and complex overlapping objects; for product photos and portraits on any distinct background it is production-grade. Check the preview at 100% zoom on the edge regions before exporting.
No - the segmentation model runs in your browser via ONNX WebAssembly. The first run downloads the model (tens of MB) to your browser cache; every photo after that is processed entirely on your device with zero upload. For photos of people, documents, and anything private, that local guarantee is the reason to prefer this over the upload-to-a-server background removers.
High detail (BiRefNet-lite, ~115MB) runs on WebGPU and is the strongest option for hair, fur, and product edges. Portrait (MODNet, ~7MB) runs on the CPU, loads in seconds, and is tuned for people. If your device has no WebGPU, the tool automatically uses MODNet; for product silhouettes on such a device, the high-detail model is worth switching devices for.
Transparent PNG, always - it is the format that carries the alpha channel every design tool, website, and marketplace expects. JPEG cannot store transparency and will put a white box behind your subject. Export at the original resolution (the model works at its input resolution and upscales the mask cleanly), and the cutout drops into overlays, thumbnails, and product listings without further work.
It uses BiRefNet-lite (MIT, ~115MB) on WebGPU for fine detail, with MODNet (Apache-2.0, ~7MB) as a CPU fallback for portraits. Both produce pixel-level alpha masks that separate subjects from backgrounds with hair-strand accuracy. No cloud API calls—unlike remove.bg which charges $0.20-$1.00 per image and uploads your photos to their servers.
First use downloads the 80MB model (cached in your browser for instant subsequent loads). Processing a single 4K image takes 2-5 seconds on a GPU-accelerated browser. Output is a transparent PNG preserving fine details like hair, fur, and semi-transparent objects. Batch mode processes multiple images sequentially without re-downloading.
This tool is also known by these tasks — each link opens the same tool with a focused guide:
What do you call a bee that's having a bad hair day?
No signups, no data sold. Every tool is free to use — the free tier allows 5 downloads or saves per day, and the optional Pro plan ($7/mo, $59/yr) adds unlimited downloads, batch processing, white-label exports and an ad-free experience.
☕Support me on Ko-fi— support the free tier100% of proceeds go towards hosting & building more free tools.