BRIA AI

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.

• https://huggingface.co/api/models/briaai/Fibo-Edit-RMBG recorded 2026-09-27

  Hub record for briaai/Fibo-Edit-RMBG: base_model briaai/Fibo-Edit, library diffusers, license_link to CC BY-NC 4.0, gated "auto".

• https://huggingface.co/api/models/briaai/RMBG-2.0 recorded 2026-09-27

  Hub record for briaai/RMBG-2.0: gated "auto", license_link to CC BY-NC 4.0, gate text "Bria AI Model weights are open source for non commercial use only".

• https://huggingface.co/briaai/VRMBG-3.0 recorded 2026-09-27

  The newer VRMBG-3.0 is a video model: "BRIA Video Background Removal v3.0" and "For still-image background removal, see RMBG-2.0."

• https://raw.githubusercontent.com/Bria-AI/RMBG-2.0/dev/LICENSE recorded 2026-09-27

  LICENSE is the Creative Commons "Attribution-NonCommercial 4.0 International" text.

• https://raw.githubusercontent.com/Bria-AI/RMBG-2.0/dev/README.md recorded 2026-09-27

  README: "The model is released under a CC BY-NC 4.0 license for non-commercial use." and "Commercial use is subject to a commercial agreement with BRIA."; training on "over 15,000 high-quality, high-resolution, manually labeled (pixel-wise accuracy), fully licensed images", with category distribution tables and no dataset release.

• https://ungh.cc/repos/Bria-AI/RMBG-2.0/files/dev recorded 2026-09-27

  Repository tree of 11 files: birefnet.py, BiRefNet_config.py, config.json, preprocessor_config.json, README, LICENSE and images; no training code.

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.

• https://huggingface.co/api/models?author=briaai&sort=downloads&limit=100&expand[]=downloads&expand[]=cardData&expand[]=gated recorded 2026-09-27

  briaai listing with RMBG-2.0 at 528,171 and RMBG-1.4 at 317,948 downloads in the trailing 30 days.

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.

• https://raw.githubusercontent.com/Bria-AI/RMBG-2.0/dev/README.md recorded 2026-09-27

  README: RMBG v2.0 is "designed to effectively separate foreground from background in a range of categories and image types" and "is developed on the BiRefNet architecture enhanced with our proprietary dataset and training scheme".