Remove Background
A neural network that runs on your device, not our servers. Drop a photo of a person, product, pet or logo and get a clean cut-out at full resolution — transparent or on the colour you choose.
Background
Saves as PNG with transparency.
About the model
ISNet (DIS) general-use segmentation, Apache-2.0, quantised to int8 (44 MB). Runs in a Web Worker with ONNX Runtime on your CPU. Inference at 1024×1024, result composited at your image's resolution (up to 4096px; Pro: 8192px). The model is cached on your device after the first download.
What this tool does
It separates the subject of a photo from its background and gives you either a transparent PNG or the subject placed on a solid colour. The work is done by a segmentation model — a small neural network — that the page downloads from ai-pix.co.uk once and then runs entirely on your computer or phone. There is no upload step, no queue, and no per-image charge, because there is no server doing the work.
This is the first AI-Pix tool that genuinely earns the "AI" in the name. We have deliberately not called the compressor or resizer "AI" — they are plain image encoding — and we will keep that distinction.
When it is the right tool
- Product photos for Amazon, eBay, Etsy, Vinted or your own shop — pick a white background and download a JPEG that meets the marketplace spec.
- Headshots and team pages — cut out on transparent, place on your brand colour, export once per person without a designer.
- Logos from screenshots or scans — the "Crisp" edge setting and "Crop to subject" give you a clean transparent mark; hand it to the favicon tool next.
- Anything confidential — unreleased products, client assets, photos of your kids. Nothing leaves the device.
How it works under the hood
- Your browser resizes a copy of the image to 1024×1024 and normalises the pixels.
- A Web Worker runs the ISNet model with ONNX Runtime (WebAssembly, multi-threaded where the browser allows) and returns a soft mask — a probability per pixel that it belongs to the subject.
- The mask is resampled back to your image's real resolution, optionally feathered or hardened, and applied as the alpha channel. If you chose a colour, the cut-out is composited onto it.
- The result is encoded on your device and offered as a download. Original and result never touch the network.
Limits and honesty
- Fine hair, fur and transparent objects are harder; use "Soft" and expect to touch up in an editor for critical work.
- Output is capped at 4096 px on the long side to keep memory sane on phones; inference is always at 1024², so extremely fine detail on very large images is interpolated.
- Very old devices without WebAssembly SIMD will be slow (30 s+). The page tells you how many threads it is using.
- The model is general-purpose. It picks the most salient subject; busy scenes with several equal subjects may keep more than you want.
Frequently asked questions
Is my photo uploaded to remove the background?
No. This is the difference between AI-Pix and remove.bg, Canva or Photoroom: those send your photo to their servers and charge per image. Here a 44 MB neural network is downloaded from this site once, stored in your browser's cache, and runs on your own CPU inside a Web Worker. Your photo never leaves the device — turn your Wi-Fi off after the model has loaded and it still works.
Is it free? How many images can I do?
Free, no account, no credits, no watermark, full resolution up to 4096 px on the long side. Because nothing runs on our servers there is nothing to meter. Do as many as you like.
How good is it compared with remove.bg?
For people, products, pets, cars, furniture and logos the result is comparable: clean edges, full-resolution output. Fine hair, fur and semi-transparent objects (glass, veils, smoke) are where cloud services with larger models still have an edge — use the "Soft" edge setting for those. The model is ISNet, the same family used by several commercial background removers.
Why does it download 44 MB the first time?
That is the model itself — the "AI" is a file of learned weights and it has to be on your device to run there. It is fetched from ai-pix.co.uk (not a third-party CDN), cached by your browser, and not downloaded again until we ship a new version. On a typical connection it takes 10–40 seconds once; after that each image takes 1–6 seconds depending on your device.
Can I put a white or coloured background behind the subject?
Yes — choose White, Black or a custom colour. That exports a JPEG on a solid background, which is what Amazon, eBay and most marketplaces want for product photos. Choose None for a transparent PNG you can drop onto anything.
Does it work on my phone?
Yes on recent phones; the first load is heavier on mobile data (44 MB) and processing takes longer than on a laptop — typically 5–15 seconds per image. If the page reports that the model could not load, your browser may be in private mode with storage disabled; it will still work, it just re-downloads next time.
What is the model and who made it?
ISNet from the "Dichotomous Image Segmentation" research project (Qin et al., 2022), released under the Apache-2.0 licence, converted to ONNX and quantised to 8-bit by us for size. It runs with Microsoft's ONNX Runtime compiled to WebAssembly. No data from your use is collected or sent back.