BRIA RMBG or BiRefNet: which to use for background removal
The two are closely related: RMBG-2.0 is built on BiRefNet, both are around 220M, both run on a CPU, and both handle soft edges, hair and transparency. The difference is in usage rights. BRIA RMBG ships under BRIA's non-commercial license (RMBG-2.0 is CC BY-NC 4.0) and business use requires a paid agreement with BRIA; in exchange the line includes the VRMBG-2.0 and VRMBG-3.0 video versions (2026-05). BiRefNet from Nankai University is MIT with no conditions, with versions for 2K and for hair and semi-transparent edges. If you need a stream of product photos in a commercial project without license negotiations, take BiRefNet.
Comparison based on catalog data
| Parameter | BRIA RMBG | BiRefNet |
|---|---|---|
| Category | Photo editing, Video | Photo editing, Computer vision |
| Developer | BRIA AI, Israel | Nankai University, China |
| Releases | Dec 2023 – May 2026 | Mar 2024 – Jul 2025 |
| Sizes | 44M – 220M | about 220M (lightweight lite versions available) |
| Hardware | Laptop | Laptop |
| Commercial use | Non-commercial only | Commercial use allowed |
| License | BRIA non-commercial license (RMBG-2.0 — CC BY-NC 4.0); for business — paid BRIA license | MIT |
| Russian | Not applicable | Not applicable |
| Ollama | No | No |
| Without GPU | Yes | Yes |
| Tasks |
|
|
Choose BRIA RMBG if
- You need background removal in video: the line includes VRMBG-2.0 and VRMBG-3.0
- It matters to you that the model was trained on licensed photos
- You are ready to sign a paid BRIA agreement, or your use is non-commercial
Choose BiRefNet if
- A commercial project with no license negotiation: BiRefNet is MIT
- Bulk product photos for marketplaces, masks for design and print
- You need 2K output and clean hair and semi-transparent edges
Other comparisons
- Wan or LTX-Video: which to choose for business
- YOLO or RF-DETR: which to choose for business
- Florence-2 or Grounding DINO: which to choose for object detection
- YOLO or Grounding DINO: which detector for video analytics
- Real-ESRGAN or SUPIR: which to use for upscaling photos
- Stable Diffusion or Kandinsky: which image model to choose
- HunyuanVideo or Wan: which open video model to choose
- Qwen or GigaChat: which to choose for business
- Qwen or Llama: which to choose for business
- GigaChat or YandexGPT: which to choose for business
- DeepSeek or Qwen: which to choose for business
- Gemma or Llama: which to choose for business
Need a model for your task?
An open model can run on your own server: data stays in-house, there is no per-request fee, and the model can be fine-tuned on your documents.
- SelectThe model and size for your task and hardware budget
- DeployOn your server or in a closed network, with an API
- Fine-tuneOn your data, or connect a knowledge base
- IntegrateInto your CRM, ERP, bot, website or team chat