VectorSpaceLab
lab · Hong Kong SAR ChinaScores
1 product on the map — 1 closed.
Openness
2 medium confidence- weights
- open (the OmniGen2 checkpoint, including its bundled Qwen2.5-VL-3B component, downloads from the Hub without a gate)
- data
- partial (the X2I2 training datasets are released under Apache 2.0; the data construction pipeline is not yet)
- code
- partial (inference code plus full-parameter and LoRA fine-tuning scripts; the pretraining pipeline is not published)
- license
- Apache-2.0 (the OmniGen2 transformer weights and the repository code)+Qwen-Research-License (the Qwen2.5-VL-3B-Instruct model shipped in the checkpoint's mllm/ folder, which every OmniGen2 run loads: non-commercial use only)
OmniGen2's own weights and code are Apache 2.0 and its training datasets are published, but the checkpoint ships Alibaba's Qwen2.5-VL-3B as its understanding model, and that component is licensed for research and evaluation only. Since every OmniGen2 run needs it, the model as distributed cannot be used commercially.
- https://huggingface.co/api/datasets?author=OmniGen2&limit=100 recorded 2026-11-12
Dataset listing for the OmniGen2 account: X2I2 (license apache-2.0, gated false) and the OmniContext benchmark.
- https://huggingface.co/api/models/OmniGen2/OmniGen2?expand[]=downloads&expand[]=likes&expand[]=cardData&expand[]=gated&expand[]=lastModified&expand[]=createdAt&expand[]=siblings recorded 2026-11-12
Hugging Face API record for OmniGen2/OmniGen2: license apache-2.0, gated false; files include mllm/model-0000N-of-00004.safetensors beside transformer/ and vae/.
- https://huggingface.co/api/models/Qwen/Qwen2.5-VL-3B-Instruct?expand[]=cardData&expand[]=gated recorded 2026-11-12
Hugging Face API record for Qwen/Qwen2.5-VL-3B-Instruct: license_name qwen-research.
- https://huggingface.co/OmniGen2/OmniGen2/raw/main/mllm/config.json recorded 2026-11-12
OmniGen2 mllm/config.json: "Qwen2_5_VLForConditionalGeneration", hidden_size 2048, 36 layers, the Qwen2.5-VL-3B shape.
- https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct/raw/main/LICENSE recorded 2026-11-12
License body: "Qwen RESEARCH LICENSE AGREEMENT"; "'Non-Commercial' shall mean for research or evaluation purposes only"; rights granted "FOR NON-COMMERCIAL PURPOSES ONLY".
- https://raw.githubusercontent.com/VectorSpaceLab/OmniGen2/main/docs/FINETUNE.md recorded 2026-11-12
Fine-tuning guide: "a training script that supports multi-GPU and multi-node distributed training"; "Both full-parameter fine-tuning and LoRA" from the released checkpoint.
- https://raw.githubusercontent.com/VectorSpaceLab/OmniGen2/main/LICENSE recorded 2026-11-12
Repository LICENSE body: Apache License, Version 2.0.
- https://raw.githubusercontent.com/VectorSpaceLab/OmniGen2/main/README.md recorded 2026-11-12
README news: "2025-06-16: We release OmniGen2"; "2025-06-30: Training code is available, see fine-tuning"; "2025-07-05: Training datasets X2I2 are available"; TODO "[ ] Training data construction pipeline"; "Inherits ... from its Qwen-VL-2.5 foundation".
Adoption
1 high confidenceAdoption is measured as Hugging Face downloads of the OmniGen2 and OmniGen2-RL checkpoints on the project's Hub account; the EditScore reward models there, and OmniGen v1 on a personal account, are left out.
- https://huggingface.co/api/models?author=OmniGen2&sort=downloads&direction=-1&limit=100 recorded 2026-11-12
Listing sorted by downloads: OmniGen2 2402, OmniGen2-EditScore7B 61, OmniGen2-EditScore7B-v1.1 58, OmniGen2-RL 47; 2,449 for the two generation checkpoints.
Capability
2 medium confidenceOmniGen2 is near the bottom of the public image arena, which scores text-to-image only and not the editing and subject-driven generation the model was built for. FLUX.2 [dev] places well above it.
- https://artificialanalysis.ai/image/leaderboard/text-to-image recorded 2026-11-12
Overall board (112 models): "OmniGen V2" 111th, Elo 703, from VectorSpaceLab, with an open-weights link to OmniGen2/OmniGen2.