Background removal models (ONNX)
ONNX conversions of open background-removal models. Each folder contains one model, its original license and a description of the conversion. All models keep their original licenses and authors; this repository only changes the file format.
| Model | Year | Author | License | Input |
|---|---|---|---|---|
| FeyNoBg | 2026 | Feyn Labs | Apache-2.0 | fixed 1024 × 1024 |
| BEN2 (base) | 2025 | Prama LLC | MIT | fixed 1024 × 1024 |
| BiRefNet Dynamic | 2025 | Peng Zheng et al. | MIT | H, W: multiples of 64, from 256 to 2304 (keep aspect ratio) |
| BiRefNet HR | 2025 | Peng Zheng et al. | MIT | fixed 2048 × 2048 |
| BiRefNet HR Matting | 2025 | Peng Zheng et al. | MIT | fixed 2048 × 2048 |
| BiRefNet Massive | 2024 | Peng Zheng et al. | MIT | fixed 1024 × 1024 |
| BiRefNet General | 2024 | Peng Zheng et al. | MIT | fixed 1024 × 1024 |
| BiRefNet Portrait | 2024 | Peng Zheng et al. | MIT | fixed 1024 × 1024 |
| BiRefNet Lite | 2024 | Peng Zheng et al. | MIT | fixed 1024 × 1024 |
| BiRefNet Lite 2K | 2024 | Peng Zheng et al. | MIT | fixed 1440 (H) × 2560 (W) |
| ISNet General | 2022 | Xuebin Qin et al. | Apache-2.0 | fixed 1024 × 1024 |
| ISNet Anime | 2022 | SkyTNT | Apache-2.0 | fixed 1024 × 1024 |
| U2-Net Human | 2020 | Xuebin Qin et al. | Apache-2.0 | fixed 320 × 320 |
| U2-Net Cloth | 2020 | Levin Dabhi | MIT | fixed 768 × 768 |
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support