whatmodel.sh / image / Lumina-Image 2.0
Lumina-Image 2.0
↗ share
+ contribute
Lumina-Image 2.0 is an open-weights image model from Alpha-VLLM (Shanghai AI Lab), released January 2025. Its published size is 2.6B parameters. Licence: apache-2.0.
Model information sources
Release date ↗ · Weights licence ↗ · Parameter count ↗
Diffusion transformer (Unified Next-DiT) only; Gemma-2-2B text encoder and VAE listed separately. The makers paper (arXiv 2503.21758) also says "with only 2.6B parameters" (Table: Lumina-Image 2.0 2.6B). The HF card rounds it: "Lumina-Image-2.0 is a 2 billion parameter flow-based diffusion transformer".
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of Lumina-Image 2.0, out of 197 with a score, then more from Alpha-VLLM (Shanghai AI Lab).
| # | model | score | where | ||
|---|---|---|---|---|---|
| 129 |  UNO | 43±4 | local | ||
| 130 |  ChronoEdit-14B | 43±3 | local | ||
| 131 | Lumina-Image 2.0 | 43±4 | api $0.075/image +local | ||
| 132 |  BLIP3o-NEXT-GRPO-Text-3B | 42±3 | comparison only | ||
| 133 |  BAGEL | 42±1 | api $0.10/image +local | ||
| more from Alpha-VLLM (Shanghai AI Lab) | |||||
| 172 | Lumina-DiMOO | 32±4 | local | ||
$ what it's good at
text to image▂▅▅▃▄▄█▃▁▂▁▁ #82 of 126 * · 94% of its score · 1 test
Beats Latent Diffusion text2img-large (LDM-KL-8-G, LAION-400M), VQ-Diffusion (ITHQ), FLUX.1 Kontext [dev] here, though they rank higher overall. Trails Playground v2 (1024px aesthetic) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| TIIF-Bench v3 (short prompts, TIIF-Evaluator) | 71.2% ✎ | ▂█····▄▄▄··▂ beats 2/11 | 94% |
$ where to run it
Through an API
api · fal.ai · $0.075/image · fal.ai ↗ · price sourceUSD for one 1-megapixel output: fal bills fal-ai/lumina-image/v2 (Lumina Image 2.0) $0.075 per output megapixel (price = rate x 1 MP; fal's rounding for 1024x1024 ~1.05 MP is not stated; larger outputs cost proportionally more). Product: fal.ai serverless text-to-image. API model: fal-ai/lumina-image/v2. Settings: 1 MP output, one image. Price basis: published per-megapixel list rate x 1 MP. Currency: USD. Not an Alpha-VLLM direct tariff. fal.ai official model page, observed 2026-09-26.
Other ways to run it
- Run it as published · needs ~13.2 GB, NVIDIA only · from huggingface.co/Alpha-VLLM (official)
uv run --with torch --with diffusers --with transformers --with accelerate python -c "import torch; from diffusers import Lumina2Pipeline; p=Lumina2Pipeline.from_pretrained('Alpha-VLLM/Lumina-Image-2.0', torch_dtype=torch.bfloat16); p.enable_model_cpu_offload(); p('A serene photograph of golden sunlight reflecting on a vast expanse of water', height=1024, width=1024, guidance_scale=4.0, num_inference_steps=50, cfg_trunc_ratio=0.25, cfg_normalization=True, generator=torch.Generator('cpu').manual_seed(0)).images[0].save('lumina_demo.png')"
Official weights and documentation: huggingface.co/Alpha-VLLM ↗
$ how the score was worked out · TIIF-Bench: How Does Your T2I… 94%
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 |
|---|---|---|
 TIIF-Bench: How Does Your T2I Model Follow Your Instructions? (arXiv v3) | 1 | 94% |
| prior · pull toward the middle | 6% | |
| weighted average | = 42.7 |
$ 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