By Andrew Armstrong 
Removing the background from a photograph is one of those jobs that ought to be free and mostly is not. remove.bg will do it in your browser, but the full-resolution result is behind credits. The desktop alternatives either want a subscription or want you to draw the mask yourself. And the models that actually do the work — U²-Net and IS-Net — have been openly licensed for years.
CutGeek is those models in a Windows application. Drop a photograph in, get a copy back with the background removed, at exactly the size the photograph was taken at. No account, no credits, no watermark, and nothing is uploaded.
Full resolution, and why that is not a favour
This is the part worth being precise about, because it is the thing the web services charge for.
The network never looks at your photograph at full size. It looks at a 320 or 1024 pixel square, and produces a mask at that size. That mask is then scaled back up and applied to the original pixels. The expensive part never depended on the resolution in the first place — so charging for a full-size download is charging for something that cost nothing extra to produce. CutGeek simply does not do that.
What it does
- Transparent PNG, a flat colour from six presets including passport blue and green screen, or the original background blurred behind the subject.
- JPEG, PNG, WebP, BMP, GIF and HEIC in; PNG or JPEG out.
- A real queue — drop in a folder and walk away, with the finished cutout shown as each one lands.
- EXIF orientation is applied. Phone photographs are very often stored sideways with a tag saying which way up they go; a decoder that ignores that tag produces a rotated cutout and looks broken. CutGeek reads it.
- The source’s own transparency is respected — an already-transparent PNG does not become opaque just because the subject mask says so.
Four models, and a pinned hash on every one
- Quick — 4.6 MB. A twentieth of the size and several times faster. Fine for a clear subject on a plain background.
- Standard — 176 MB. The usual choice. Start here.
- People — 176 MB. Trained on people; noticeably better on portraits, and only on portraits.
- Detailed — 179 MB. Reads the image at 1024 pixels instead of 320, so hair and thin edges survive. Slower.
Nothing is shipped in the installer — you pick one on the Models screen and it is fetched once and kept. Every download is checked against a SHA-256 recorded inside the application and deleted if it does not match. That is not ceremony: these weights come from somebody else’s GitHub release and are executed by the ONNX runtime, and a release asset can be replaced by whoever owns the repository. Pinning the hash means CutGeek runs the exact file it was tested with, or it runs nothing at all. The hashes are shown on the Models screen so you can see them.
What it will not do
- It does not upload your photographs. The model runs in the application, on your own processor. There is no account and no server.
- It does not shrink the result or watermark it. The cutout is the size of the photograph that went in.
- It does not change your originals. The photograph is only ever read, and if a cutout of that name is already there the next one is numbered.
- It does not count anything. No credits, no monthly allowance, no per-image limit, no trial.
Requirements
- Windows 10 version 1809 or later, 64-bit. Windows 11 fully supported.
- Runs as a normal user — no administrator rights are needed or requested.
- About 76 MB for the installer. .NET 8 is inside it; there is nothing to install first.
- CPU only. There is no GPU package, because it would pull in several hundred megabytes of CUDA that most people cannot use.
- GPL-3.0. Free to use, including at work. No paid tier, ever.
Where to get it
The installer and a portable build are on the CutGeek product page, with SHA-256 checksums for both. The source is on GitHub.
A manual brush for correcting the edge is the next thing planned. If you hit a bug or want something adding, open an issue and it will be read.
Discover more from TechyGeeksHome
Subscribe to get the latest posts sent to your email.
