omnigen2-base-df5dca8a
Mirror of the OmniGen2 unified multimodal model weights in diffusers layout, pinned to one upstream revision.
What it is for
The repo holds a copy of OmniGen2: the Qwen2.5-VL based MLLM, its processor, the OmniGen2 diffusion transformer and its code, the VAE and the flow-match Euler scheduler.
OmniGen2 has separate decoding paths for text and images. It does visual understanding, text-to-image generation, instruction-guided image editing and in-context generation.
The repo also keeps upstream's example images under assets/ and a CITATION.cff.
Provenance
- Upstream source: OmniGen2/OmniGen2. The suffix
df5dca8ain the repo name is the upstream revision. - Code: VectorSpaceLab/OmniGen2.
- Paper: OmniGen2: Exploration to Advanced Multimodal Generation.
Use
hf download chibifire/omnigen2-base-df5dca8a --local-dir omnigen2
Load the folder with the pipeline in the upstream code repo. See the upstream card for setup, examples and tuning parameters.
Licence
Apache 2.0, as upstream licenses it. See the Apache License 2.0 terms and the upstream card.