Latent Consistency Models

Official Repository of the paper: Latent Consistency Models.

Project Page: https://latent-consistency-models.github.io

Try our Hugging Face demos:

Model Descriptions:

Distilled from Dreamshaper v7 fine-tune of Stable-Diffusion v1-5 with only 4,000 training iterations (~32 A100 GPU Hours).

Generation Results:

By distilling classifier-free guidance into the model's input, LCM can generate high-quality images in very short inference time. We compare the inference time at the setting of 768 x 768 resolution, CFG scale w=8, batchsize=4, using a A800 GPU.

Usage

You can try out Latency Consistency Models directly on:

To run the model yourself, you can leverage the 🧨 Diffusers library:

1. Install the library:

pip install --upgrade diffusers  # make sure to use at least diffusers >= 0.22
pip install transformers accelerate

2. Run the model:

from diffusers import DiffusionPipeline
import torch

pipe = DiffusionPipeline.from_pretrained("SimianLuo/LCM_Dreamshaper_v7")

# To save GPU memory, torch.float16 can be used, but it may compromise image quality.
pipe.to(torch_device="cuda", torch_dtype=torch.float32)

prompt = "Self-portrait oil painting, a beautiful cyborg with golden hair, 8k"

# Can be set to 1~50 steps. LCM support fast inference even <= 4 steps. Recommend: 1~8 steps.
num_inference_steps = 4 

images = pipe(prompt=prompt, num_inference_steps=num_inference_steps, guidance_scale=8.0, lcm_origin_steps=50, output_type="pil").images

For more information, please have a look at the official docs: 👉 https://huggingface.co/docs/diffusers/api/pipelines/latent_consistency_models#latent-consistency-models

Usage (Deprecated)

1. Install the library:

pip install diffusers transformers accelerate

2. Run the model:

from diffusers import DiffusionPipeline
import torch

pipe = DiffusionPipeline.from_pretrained("SimianLuo/LCM_Dreamshaper_v7", custom_pipeline="latent_consistency_txt2img", custom_revision="main", revision="fb9c5d")

# To save GPU memory, torch.float16 can be used, but it may compromise image quality.
pipe.to(torch_device="cuda", torch_dtype=torch.float32)

prompt = "Self-portrait oil painting, a beautiful cyborg with golden hair, 8k"

# Can be set to 1~50 steps. LCM support fast inference even <= 4 steps. Recommend: 1~8 steps.
num_inference_steps = 4 

images = pipe(prompt=prompt, num_inference_steps=num_inference_steps, guidance_scale=8.0, output_type="pil").images

BibTeX

@misc{luo2023latent,
      title={Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference}, 
      author={Simian Luo and Yiqin Tan and Longbo Huang and Jian Li and Hang Zhao},
      year={2023},
      eprint={2310.04378},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}