Files changed (1) hide show
  1. README.md +64 -0
README.md CHANGED
@@ -1,3 +1,67 @@
1
  ---
2
  license: mit
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: mit
3
+ language:
4
+ - en
5
+ pipeline_tag: text-to-image
6
+ tags:
7
+ - text-to-image
8
  ---
9
+
10
+ # Latent Consistency Models
11
+
12
+ Official Repository of the paper: *[Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference](https://arxiv.org/abs/2310.04378)*.
13
+
14
+ Project Page: https://latent-consistency-models.github.io
15
+
16
+
17
+ <p align="center">
18
+ <img src="teaser.png">
19
+ </p>
20
+
21
+ 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.
22
+
23
+ <p align="center">
24
+ <img src="speed_fid.png">
25
+ </p>
26
+
27
+ ## Usage
28
+
29
+ You can try out Latency Consistency Models directly on:
30
+ [![Hugging Face Spaces](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-blue)](https://huggingface.co/spaces/SimianLuo/Latent_Consistency_Model)
31
+
32
+ To run the model yourself, you can leverage the 🧨 Diffusers library:
33
+ 1. Install the library:
34
+ ```
35
+ pip install diffusers transformers accelerate
36
+ ```
37
+
38
+ 2. Run the model:
39
+ ```py
40
+ from diffusers import DiffusionPipeline
41
+ import torch
42
+
43
+ pipe = DiffusionPipeline.from_pretrained("SimianLuo/LCM_Dreamshaper_v7", custom_pipeline="latent_consistency_txt2img")
44
+
45
+ # To save GPU memory, torch.float16 can be used, but it may compromise image quality.
46
+ pipe.to(torch_device="cuda", torch_dtype=torch.float32)
47
+
48
+ prompt = "Self-portrait oil painting, a beautiful cyborg with golden hair, 8k"
49
+
50
+ # Can be set to 1~50 steps. LCM support fast inference even <= 4 steps. Recommend: 1~8 steps.
51
+ num_inference_steps = 4
52
+
53
+ images = pipe(prompt=prompt, num_inference_steps=num_inference_steps, guidance_scale=8.0, lcm_origin_steps=50, output_type="pil", custom_revision=main).images
54
+ ```
55
+
56
+ ## BibTeX
57
+
58
+ ```bibtex
59
+ @misc{luo2023latent,
60
+ title={Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference},
61
+ author={Simian Luo and Yiqin Tan and Longbo Huang and Jian Li and Hang Zhao},
62
+ year={2023},
63
+ eprint={2310.04378},
64
+ archivePrefix={arXiv},
65
+ primaryClass={cs.CV}
66
+ }
67
+ ```