Background Remover
Remove image backgrounds automatically with AI.
Uses an on-device AI model (downloaded once on first use, ~40MB). Your image is never uploaded.
Choose an image.
What is Background Remover?
Background Remover cuts the subject out of a photo automatically with an AI model that runs on your device, producing a transparent PNG. It's a free way to remove an image background online — even tricky edges like hair — with the model downloaded once and cached in your browser.
Key features
- AI subject cutout that handles complex edges like hair
- Outputs a transparent PNG with a checkerboard preview
- Model downloads once, then is cached for later runs
- Works best on clear subjects: people, products, animals
How to remove an image background with AI
Automatic background removal usually means either a fiddly manual selection in an image editor or uploading your photo to a website that processes it on a server. This tool does it with a neural segmentation model that runs inside your browser — the photo is never uploaded, and there's no per-image size cap, because nothing goes to a server in the first place.
What it does well
The model is trained to separate a foreground subject from everything behind it, so it handles the edges that defeat simple tools: strands of hair, fur, the gap between an arm and a torso. It works best when there's one clear subject — a person, a product on a surface, a pet. Photos with several overlapping subjects, or a subject that blends into a background of the same color, give the model less to lock onto, and the cutout gets rougher.
Using it
- Choose an image. A preview of the original appears immediately.
- Remove background & download. On the very first run the tool downloads the model file once. You'll see a downloading percentage, then a processing stage while the model runs. When it finishes, the result is shown on a checkerboard (that's just how transparency is displayed) and saved as
yourname_no-bg.png.
After that first download the model is cached in your browser, so every later cutout skips straight to processing.
Why it's slower the first time — and on big images
Two things cost time. The one-time model download is a few megabytes; on a slow connection the first run feels sluggish, then later runs are quick. The second cost is inference itself: because the network runs on your own CPU/GPU through WebAssembly, a large image (say a 12-megapixel phone photo) takes noticeably longer than a small one, and on an older phone it can take several seconds. Downscaling a huge photo first with Image Resize speeds things up when you don't need full resolution.
When to use a simpler tool instead
If your subject already sits on a solid white or near-white background — a product shot, a scanned logo, a signature — the Remove White Background tool is faster and more predictable. It uses a plain color threshold rather than a model, so there's no download and no guessing; it simply turns matching pixels transparent. Reach for the AI remover when the background is a real scene that a color threshold can't separate cleanly.
The output
The result is always a transparent PNG, the only common format that stores per-pixel transparency (a JPG can't — it would fill the cut-out area with a solid color). Hair and soft edges keep a feathered alpha channel, which is what makes a composited result look natural instead of scissored out. If you need the subject on a specific colored backdrop, drop the PNG onto that color afterward, or place it in a layout where the transparency does its job.
Frequently asked questions
- Does my photo get uploaded to a server?
- No. The segmentation model runs on your device through WebAssembly, so the image never leaves your browser. That's also why there's no file-size limit like server tools impose; the only ceiling is your device's memory.
- Why is the first run so slow, then fast after that?
- The first cutout downloads the model file once (a few megabytes) before it can run, shown as a "downloading" percentage. After that it's cached in your browser, so every later image skips the download and goes straight to the processing stage.
- The edges around hair look rough — how do I improve them?
- The model does best with one clear subject against a background that contrasts with it. A subject that blends into a same-colored background gives it less to separate, so choosing or shooting a photo with more contrast between subject and background produces cleaner hair edges.
- What format is the result, and can I get a JPG?
- It's always a transparent PNG, because JPG can't store transparency — a JPG would fill the cut-out area with a solid color. Keep the PNG if you need the transparency; convert it later with Image Converter if you specifically need a JPG on a solid background.
- My background is plain white — is this the right tool?
- For a solid white or near-white backdrop, Remove White Background is faster: it uses a color threshold instead of an AI model, so there's no model download and the result is more predictable. Use the AI remover when the background is a real scene a threshold can't separate.
Privacy
The segmentation model runs on-device via WebAssembly after a one-time download; your photo is never uploaded, which is also why there's no server-side size limit.