How to remove the background from a photo (in your browser, no Photoshop)

Cutting out the subject from a photo used to be a Photoshop skill. Pen tool tracing around the edges, refining the mask, painting in the strands of hair that got missed. A 30-minute job for a clean result. The good news is that AI-based background removal has gotten very good in the last few years; the better news is that it now runs entirely in a browser tab. Drop a photo, click a button, get a transparent PNG.

How AI background removal works

The underlying technology is a neural network — specifically a U²-Net-style architecture trained on roughly 100,000 images of subjects against backgrounds. The model learned to predict, for every pixel in an image, whether it belongs to "the subject" or "the background". When you run an image through it, you get back a mask (a black-and-white image where the subject is white and the background is black). Apply the mask to the original photo and you get a cutout with transparent background — which you save as a PNG.

Until recently, the models were hundreds of MB and required GPU compute. They lived on cloud servers and were exposed via APIs (remove.bg, Adobe's Background Remover). The change: ONNX Runtime Web — a JavaScript port of Microsoft's neural-network inference engine — can run these models on the browser's GPU via WebGL or WebGPU. The model is downloaded once (~70 MB compressed) and cached. After that, each photo processes in 5–20 seconds entirely in your browser.

Step-by-step with our tool

1. Open justfiletools.com/tools/background-remover.
2. Drop a photo. JPG, PNG, or WebP all work.
3. Wait for the model to load (~70 MB, one-time per session).
4. Watch the progress indicator as the cutout is generated. 5–20 seconds depending on image size and your machine.
5. Download the resulting PNG. The background is transparent; the subject is cleanly cut out.

Quality expectations

For clean subject-against-distinct-background photos, results are excellent. A person against a wall, a product against a sweep, a pet against grass — these produce crisp edges, accurate hair boundaries, no visible artifacts. About 80% of casual phone photos fall here.

For slightly tricky cases (subject partially obscured, similar-color background, complex lighting), the cutout works but with occasional edge artifacts. Sometimes a small region of background gets included as "subject" or vice versa. Usually fixable with a quick touch-up in any image editor.

For hard cases (glass, hair flying in motion, semi-transparent edges, very busy backgrounds matching the subject's colors), expect imperfect results. Touch-up work is needed. For commercial product photography requiring pixel-perfect edges, professional tools (Photoshop's Select Subject + Refine Edge, Topaz Mask AI, cloud services with manual review) are the right tool.

Use cases this tool is right for

• Product photos for e-commerce listings. Clean cutout against transparent background lets you composite over any backdrop or place the product in a designed layout.
• Social media graphics. Pull a subject out of a photo and drop them onto a brand-colored background or in a designed layout.
• Profile pictures. Cut yourself out of a vacation photo for a LinkedIn or Twitter avatar.
• Stickers / emoji for messaging apps. Subject-only PNG works as an iMessage sticker or a Discord custom emoji.
• Quick mock-ups. Designer needs a "person on a phone" image for a marketing site mock-up — cut a stock photo subject out and composite over the designed background.

Cases this tool is wrong for

• Commercial print work requiring pixel-perfect edges. Use Photoshop or a paid cloud service with manual touchup.
• Removing one specific object from a complex scene (subject removal, not background removal). The model picks the most salient subject; for "remove the person on the left but keep the building", you need different tools (Photoshop's Content-Aware Fill, AI inpainting).
• Multiple subjects. The model usually picks one salient subject. Images with multiple subjects of similar importance can produce unpredictable results.
• Very large images. Above 4000×4000 pixels, browser memory becomes the binding constraint. Resize first to under 2048 pixels on the long side, then remove the background, then upscale the cutout if needed.

Performance details

First run: 30 seconds to 2 minutes for the ~70 MB model to download, then 5–20 seconds for the actual background removal. The model lives in your browser's IndexedDB after first download; subsequent visits to the page reuse the cached model.

Subsequent images in the same session: 5–20 seconds per image depending on resolution. The work happens on your GPU via WebGL or WebGPU. A faster GPU runs the model faster; integrated graphics work but are slower than discrete.

Memory: a 4 MP image (2400×1800) uses about 200 MB of browser tab memory during processing. 8 MP images use about 400 MB. Browsers cap tab memory around 2–4 GB, so very large images can fail. The workaround is to resize first.

Alternative approaches and when to use them

• remove.bg (web service). Pioneered the browser-friendly background-removal UX. High quality. Free tier limited to small images; paid tier for full-resolution. Uploads your photo.
• Adobe Express Remove Background. Web tool in Adobe's product suite. Free, good quality. Uploads.
• Photoshop Select Subject + Refine Edge. The professional answer. More work, better results, full control over edge softness and matting. Right tool when quality matters more than speed.
• Photoroom mobile app. Phone-native, fast, optimized for product photos. Right tool for ongoing e-commerce listing work on mobile.
• iOS native lift-from-background. On iOS 16+, long-press on a subject in any photo to lift it from the background. Free, fast, runs on-device. Best option if you're already on iOS.

Privacy considerations

Background-removal services often process millions of user photos. Your photo becomes training data for the next-generation model unless you opt out, and most services have terms that allow retention. For sensitive photos (kids, ID photos, confidential product designs), the in-browser approach is the right answer.

The neural network in our tool runs as WebAssembly entirely in your browser via ONNX Runtime Web. The model weights are fetched once from a public CDN (these are open-source weights, not your image) and cached. Your image is processed locally; the cutout is generated locally. The only network requests during operation are the initial model-weight downloads, which are public files identical for every user of the tool. Verify in DevTools — subsequent images produce zero requests.

Related tools and guides

• Background Remover — the tool this guide covers.
• Compress Image — the PNG output is large; compress for web.
• Image Watermark — add a watermark to the cutout after.
• EXIF Viewer / Stripper — clean metadata from the source photo.