whatmodel.sh / image / OmniGen (v1)
OmniGen (v1)
↗ share
+ contribute
OmniGen (v1) is an open-weights image model from BAAI, released October 2024. Its published size is 3.88B parameters. Licence: mit.
Catalogue identity notes
OmniGen v1 (Xiao et al., 'OmniGen: Unified Image Generation', BAAI / VectorSpaceLab), released 22 Oct 2024 as Shitao/OmniGen-v1 (MIT); the only OmniGen 1 checkpoint (BAAI/OmniGen-v1 is a later mirror with the same parameter count). Reported as 'OmniGen [98]' (citing Xiao et al. 2025) in BAAI's own OmniGen2 report Table 4 and as 'OmniGen' (date column 24.09.11, repository github.com/VectorSpaceLab/OmniGen) in EdiVal Table 3. Distinct from omnigen2.
Model information sources
Release date ↗ · Weights licence ↗ · Parameter count ↗
HF safetensors metadata for Shitao/OmniGen-v1@58e249c: 3,875,318,800 parameters (F32) in model.safetensors, the single transformer that includes the LLM backbone; equals the BF16 transformer of the official diffusers conversion Shitao/OmniGen-v1-diffusers@016e2f6 (7.75 GB). VAE (vae/diffusion_pytorch_model.safetensors, 0.33 GB) excluded. Observed 2026-10-09.
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of OmniGen (v1), out of 197 with a score, then more from BAAI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 167 |  Stable Diffusion XL 1.0 | 34±1 | api $0.0019/image +local | ||
| 168 |  LCM-SDXL | 34±3 | local | ||
| 169 | OmniGen (v1) | 34±4 | local | ||
| 170 | SDXL-Turbo | 34±2 | local | ||
| 171 |  IF-L | 33±3 | local | ||
| more from BAAI | |||||
| 48 | Emu3.5-Image | 60±5 | local | ||
| 69 | Emu3.5 | 57±5 | local | ||
| 136 | OmniGen2 | 42±2 | api $0.15/image +local | ||
$ what it's good at
image to image / editing▁▁·▃▂▃█▃▃▄▁▁ #59 of 62 * · 18% of its score · 1 test
| test | result | vs the others in its run | counts |
|---|---|---|---|
| OmniContext average (GPT-4.1 judge) lab | 43.4% ✎ | ▃▃▃··▃·····█ beats 1/6 | 18% |
other tests2 not in a task · 74% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| WEAVEBench in-context multi-turn editing (Avg) | 40.4% ✎ | █▄▄·▄▄▄▄█·▄█ beats 0/12 | 41% |
| EdiVal multi-turn editing overall (EdiVal-O after turn 3) | 29.9% ✎ | ▂·▂▃·▃·▃▂·▃█ beats 4/14 | 33% |
$ where to run it
Other ways to run it
- Run it as published · needs ~9 GB, NVIDIA only · from huggingface.co/Shitao
git clone https://github.com/staoxiao/OmniGen.git && cd OmniGen && pip install -e . && python -c "from OmniGen import OmniGenPipeline; pipe = OmniGenPipeline.from_pretrained('Shitao/OmniGen-v1'); pipe(prompt='A curly-haired man in a red shirt is drinking tea.', height=1024, width=1024, guidance_scale=2.5, seed=0)[0].save('example_t2i.png')"
Weights and documentation: huggingface.co/Shitao ↗
$ how the score was worked out · WEAVE: Unleashing and Benchma… 41%, EdiVal-Agent: An Object-Centr… 33%, BAAI — OmniGen2 technical rep… 18%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 WEAVE: Unleashing and Benchmarking the In-context Interleaved Comprehension and Generation (arXiv:2511.11434v1) | 1 | 41% |
| EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editing (arXiv:2509.13399v3) | 1 | 33% |
| BAAI — OmniGen2 technical report · lab-reported | 1 | 18% |
| prior · pull toward the middle | 8% | |
| weighted average | = 34.0 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json