A remove.bg alternative that never uploads your image
remove.bg popularised one-click background removal and sells it by the credit through a web app and an API. The model class underneath it is open: RMBG-1.4, MODNet and BiRefNet are published on Hugging Face with ONNX exports, and they are small enough — 25 MB to about 170 MB — to download into a browser and run on your own graphics card. This page does exactly that. Your image is read from a file input into a canvas and never leaves the tab, there is no credit counter, and the cutout you get back is the resolution you put in.
At a glance
- What replaces it
- RMBG-1.4, MODNet, BiRefNet Lite (open ONNX models)
- Where it runs
- Your GPU, in the browser tab, via WebGPU
- Image upload
- None — the file never leaves your device
- Cost per image
- Nothing; one download, then unlimited local runs
- Output resolution
- The resolution you gave it
- Works offline
- Yes, once the model is cached
- Honest limitation
- A single open checkpoint is less consistent on hair and fur than a tuned commercial pipeline
Where remove.bg is better: it is a tuned product, not a raw model. Its pipeline handles hair, fur and motion blur more consistently than a single open checkpoint does, it has a bulk API with an SLA behind it, and it works on a phone with no GPU and no download. If you are removing backgrounds from ten thousand product photos on a schedule, a paid API is the right tool and this is not.
Where local wins: the first run costs a download and every run after it costs nothing, so there is no per-image maths to do and no quota to hit. Nothing is transmitted, which matters when the images are client work, medical, legal or simply under an NDA — the usual blocker on an image-upload service is not price, it is that the picture is not yours to upload. And it keeps working with the network off.
Three models are worth trying in order. MODNet is the smallest at about 25 MB and is tuned for people and portraits. RMBG-1.4 is the general-purpose default and handles products and objects better. BiRefNet Lite is the largest and the most careful around thin structures. They disagree most on hair and on semi-transparent edges, which is the case worth testing with your own images.
remove.bg alternativefree background removerremove background without uploadrmbg-1.4 browsermodnet background removalbirefnet onlineoffline background removerprivate background removalunlimited background removal freebackground removal no watermark
Frequently asked questions
Is the image really not uploaded?
It is not. The model weights travel one way — from Hugging Face to your browser cache — and the image is read locally into a canvas. hjLabs.in runs no inference server, so there is nowhere for it to go. You can confirm it by opening the network tab, or by disconnecting after the model loads.
Is there a limit on how many images I can do?
No, because we are not paying for the compute — your GPU is. There is no credit balance, no daily cap and no watermark on the result.
How does the quality compare to remove.bg?
On clean subjects with clear edges, comparable. On hair, fur, fine foliage and semi-transparent edges a tuned commercial pipeline is still better than a single open checkpoint. Trying all three models on one difficult image is the fastest way to see the gap for your own material.
What formats can I use, and what comes back?
Anything the browser can decode — PNG, JPEG, WebP — and you get a PNG with a real alpha channel at the input resolution, so it composites straight onto another background.
Can I use the output commercially?
That depends on the individual model, not on this site. Each model page links to its Hugging Face repository, where the licence is stated — most of the catalog is Apache-2.0 or MIT, a few are research-only. We add nothing on top: hjLabs.in claims no rights over anything you produce here.