AI Potluck
Back to Gap Map Infrastructure / Classic ML & computer vision

BRIA RMBG

BRIA AI
restricted / Overall score: 3.4

Background-removal model line from BRIA AI. RMBG-2.0, the current still-image model, is built on the BiRefNet architecture and trained on images BRIA licensed, a dataset it presents as safe for commercial deployment. The weights are published for non-commercial use, and commercial use goes through a BRIA agreement or its paid background-removal API.

Openness

2 high confidence
2.0
weights
open(on the Hub behind an automatic click-through gate)
data
documented-not-released(over 15,000 licensed images, described by category
code
partial(inference code only)
license
CC-BY-NC-4.0(RMBG-2.0 weights

The RMBG-2.0 weights can be downloaded after a click-through, but only for non-commercial use under CC BY-NC 4.0; businesses need a commercial agreement with BRIA or its paid API. BRIA describes the licensed training images without releasing them, and publishes inference code only. BRIA's newer Fibo-Edit-RMBG, a background-removal fine-tune of its Fibo-Edit image editor, and the video model VRMBG-3.0 are separate models; BRIA's own VRMBG-3.0 card points still-image users to RMBG-2.0, and Fibo-Edit-RMBG is non-commercial too.

Adoption

3 high confidence
3.0

Hugging Face downloads over the trailing 30 days of RMBG-2.0; the older RMBG-1.4 has a smaller count and is not declared. BRIA's API usage is not counted.

Capability

4 medium confidence
4.0

RMBG-2.0 does the same job as BiRefNet, cutting the subject out of any image with no training, and uses its architecture; BRIA's contribution is the licensed training data. It offers no variants beyond that one task.

Verified 2026-09-27