whatmodel.sh / image / LCM-SDXL
LCM-SDXL
↗ share
+ contribute
LCM-SDXL is an open-weights image model from latent-consistency, released November 2023. Its published size is 2.57B parameters. Licence: openrail++.
Catalogue identity notes
latent-consistency/lcm-sdxl full UNet distillation (not the LCM-LoRA). Memory estimated as for SDXL 1.0.
Model information sources
UNet only: HF Hub safetensors total for the official UNet-only repo (pinned revision), 2,567,545,604 = diffusion_pytorch_model.safetensors; used with the SDXL base pipeline, whose text encoders and VAE are not included. Observed 2026-10-07.
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of LCM-SDXL, out of 197 with a score, then more from latent-consistency.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 166 |  IF-XL | 35±2 | local | ||
| 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 | ||
| more from latent-consistency | |||||
| 175 | LCM-SSD-1B | 32±4 | local | ||
$ what it's good at
text to image▂▅▅▃▄▄█▃▁▂▁▁ #109 of 126 * · 44% of its score · 1 test
Beats Latent Diffusion text2img-large (LDM-KL-8-G, LAION-400M), VQ-Diffusion (ITHQ) here, though they rank higher overall. Trails IF-L, SDXL-Turbo here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| EvalMuse-40K model ranking (FGA-BLIP2 overall alignment) | 2.77 / 5 ✎ | █▄·▄██▄▄█▄▄▄ beats 4/14 | 44% |
other tests2 not in a task · 54% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| EvalAlign human text-image alignment score | 5.1800 mean human alignment score ✎ | █▄▄·█·▄█▄▄·▄ beats 9/11 | 40% |
| MS-COCO 10K FID (whole image) ↓ (SDXL-Lightning protocol) | 21.8500 FID ✎ | █··██······█ beats 2/3 | 14% |
$ where to run it
Other ways to run it
- Run it as published · needs ~8.9 GB, NVIDIA only · from huggingface.co/latent-consistency (official)
uv run --with diffusers --with transformers --with accelerate --with torch python -c "import torch; from diffusers import UNet2DConditionModel, DiffusionPipeline, LCMScheduler; u=UNet2DConditionModel.from_pretrained('latent-consistency/lcm-sdxl', torch_dtype=torch.float16, variant='fp16'); p=DiffusionPipeline.from_pretrained('stabilityai/stable-diffusion-xl-base-1.0', unet=u, torch_dtype=torch.float16, variant='fp16'); p.scheduler=LCMScheduler.from_config(p.scheduler.config); p.to('cuda'); p('a cat holding a sign', num_inference_steps=4, guidance_scale=8.0).images[0].save('out.png')"
Official weights and documentation: huggingface.co/latent-consistency ↗
$ how the score was worked out · EvalMuse-40K: A Reliable and… 44%, EvalAlign: Supervised Fine-Tu… 40%, SDXL-Lightning: Progressive A… 14%
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 |
|---|---|---|
 EvalMuse-40K: A Reliable and Fine-Grained Benchmark with Comprehensive Human Annotations for Text-to-Image Generation Model Evaluation (arXiv v2) | 1 | 44% |
| EvalAlign: Supervised Fine-Tuning Multimodal LLMs with Human-Aligned Data for Evaluating Text-to-Image Models (arXiv:2406.16562v3) | 1 | 40% |
| SDXL-Lightning: Progressive Adversarial Diffusion Distillation (arXiv:2402.13929v3) | 1 | 14% |
| prior · pull toward the middle | 3% | |
| 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