whatmodel.sh / image / Latent Diffusion text2img-large (LDM-KL-8-G, LAION-400M)
Latent Diffusion text2img-large (LDM-KL-8-G, LAION-400M)
↗ share
+ contribute
Latent Diffusion text2img-large (LDM-KL-8-G, LAION-400M) is an open-weights image model from CompVis, released April 2022. Its published size is 1.45B parameters. Licence: apache-2.0.
Catalogue identity notes
The only officially released CompVis text-to-image LDM: the 1.45B LAION-400M model (latent-diffusion README, April 2022), packaged as CompVis/ldm-text2im-large-256.
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of Latent Diffusion text2img-large (LDM-KL-8-G, LAION-400M), out of 197 with a score, then more from CompVis.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 104 |  Runway Gen-4 Image | 50±3 | api $0.050/image | ||
| 105 |  Lucid Origin | 49±3 | api $0.017/image | ||
| 106 | Latent Diffusion text2img-large (LDM-KL-8-G, LAION-400M) | 48±15 | local | ||
| 107 |  Kandinsky 3 | 48±11 | local | ||
| 108 |  Step1X-Edit v1.1 | 48±3 | local | ||
| more from CompVis | |||||
| 186 | Stable Diffusion v1.4 | 27±2 | local | ||
| 187 | Stable Diffusion v1.3 | 27±4 | local | ||
| 188 | Stable Diffusion v1.1 | 26±3 | local | ||
$ what it's good at
text to image▂▅▅▃▄▄█▃▁▂▁▁ #116 of 126 * · 6.1% of its score · 1 test
Trails CogView2, Versatile Diffusion v1.0, Dreamlike Photoreal 2.0 here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| HPS v2 benchmark (4-style average) | 25.8% ✎ | ▃··▃▅·▅·▅▅█▅ beats 4/14 | 6.1% |
$ where to run it
Other ways to run it
- Run it as published · needs ~7.5 GB, NVIDIA only · from huggingface.co/CompVis (official)
uv run --with diffusers --with transformers --with accelerate --with torch python -c "from diffusers import DiffusionPipeline; p=DiffusionPipeline.from_pretrained('CompVis/ldm-text2im-large-256').to('cuda'); p(['a cat holding a sign'], num_inference_steps=50, eta=0.3, guidance_scale=6).images[0].save('out.png')"
Official weights and documentation: huggingface.co/CompVis ↗
$ how the score was worked out · tgxs002/HPSv2 repository READ… 6%
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 |
|---|---|---|
 tgxs002/HPSv2 repository README, HPS v2 benchmark table | 1 | 6% |
| prior · pull toward the middle | 94% | |
| weighted average | = 48.4 |
$ 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