whatmodel.sh / image / Lumina-DiMOO
Lumina-DiMOO
↗ share
+ contribute
Lumina-DiMOO is an open-weights image model from Alpha-VLLM (Shanghai AI Lab), released September 2025. Its published size is 8B parameters. Licence: apache-2.0.
Catalogue identity notes
Alpha-VLLM/Lumina-DiMOO (single released checkpoint; HF revision f014549, created 2025-09-09, Apache-2.0; technical report arXiv 2510.06308). Unified discrete-diffusion model for generation, editing and understanding. No local command recorded (runs through the authors' repository scripts).
Model information sources
8B as the makers' report lists it ('# Params.') for Lumina-DiMOO, a discrete-diffusion model initialised from LLaDA-Base without structural modifications; whether the image tokenizer/decoder is counted is not stated.
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of Lumina-DiMOO, out of 197 with a score, then more from Alpha-VLLM (Shanghai AI Lab).
| # | model | score | where | ||
|---|---|---|---|---|---|
| 170 |  SDXL-Turbo | 34±2 | local | ||
| 171 |  IF-L | 33±3 | local | ||
| 172 | Lumina-DiMOO | 32±4 | local | ||
| 173 |  UltraEdit (SD3, with mask) | 32±7 | comparison only | ||
| 174 |  VILA-U 7B (256px) | 32±7 | local | ||
| more from Alpha-VLLM (Shanghai AI Lab) | |||||
| 131 | Lumina-Image 2.0 | 43±4 | api $0.075/image +local | ||
$ what it's good at
image to image / editing▁▃·▃▂▃█▃▃▄▁▁ #60 of 62 * · 83% of its score · 1 test
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Edit-Compass (EN, overall) | 26.6% ✎ | ▃▃▂·▃▆▂█·▃·▃ beats 3/20 | 83% |
other tests1 not in a task · 9.4% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| ViGoR-Bench zero-shot visual reasoning, image models (Result Avg) | 34.0% ✎ | ▂···▂·▂█▆▅·▂ beats 10/15 | 9.4% |
$ where to run it
Other ways to run it
- Run it as published · needs ~38.9 GB, NVIDIA only · from github.com/Alpha-VLLM (official)
git clone https://github.com/Alpha-VLLM/Lumina-DiMOO.git && cd Lumina-DiMOO && conda create -n lumina_dimoo python=3.10 -y && conda activate lumina_dimoo && pip install -r requirements.txt && python inference/inference_t2i.py --checkpoint Alpha-VLLM/Lumina-DiMOO --prompt "A striking photograph of a glass of orange juice on a wooden kitchen table" --height 768 --width 1536 --timesteps 64 --cfg_scale 4.0 --seed 65513 --vae_ckpt Alpha-VLLM/Lumina-DiMOO --output_dir output/results_text_to_image
Official weights and documentation: huggingface.co/Alpha-VLLM ↗
$ how the score was worked out · Edit-Compass & EditReward-Com… 83%, ViGoR-Bench: How Far Are Visu… 9%
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 |
|---|---|---|
 Edit-Compass & EditReward-Compass: A Unified Benchmark for Image Editing and Reward Modeling | 1 | 83% |
| ViGoR-Bench: How Far Are Visual Generative Models From Zero-Shot Visual Reasoners? (arXiv:2603.25823v1) | 1 | 9% |
| prior · pull toward the middle | 7% | |
| weighted average | = 32.4 |
$ 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