| license: apache-2.0 | |
| library_name: diffusers | |
| pipeline_tag: image-to-image | |
| # OmniGen2-OpenSubject | |
| This repository provides an **open-source baseline** model, obtained by fine-tuning **OmniGen2** on the sampled 500k **OpenSubject** corpus and 100k internal T2I samples to maintain prompt-following ability. The model is optimized for **subject-driven image generation**, while maintaining general **prompt-following** ability. | |
| - **Paper**: [OpenSubject: Leveraging Video-Derived Identity and Diversity Priors for Subject-driven Image Generation and Manipulation](https://huggingface.co/papers/2512.08294) | |
| - **Code**: [https://github.com/LAW1223/OpenSubject](https://github.com/LAW1223/OpenSubject) |