license: cc-by-nc-4.0
library_name: diffusers
tags:
- text-to-image
- stable-diffusion
- diffusion distillation
DMD2 Model Card
Improved Distribution Matching Distillation for Fast Image Synthesis,
Tianwei Yin, Michaël Gharbi, Taesung Park, Richard Zhang, Eli Shechtman, Frédo Durand, William T. Freeman
Contact
Feel free to contact us if you have any questions about the paper!
Tianwei Yin tianweiy@mit.edu
Huggingface Demo
Our 4-step (much higher quality, 2X slower) Text-to-Image demo is hosted at DMD2-4step
Our 1-step Text-to-Image demo is hosted at DMD2-1step
Usage
We can use the standard diffuser pipeline:
4-step generation
import torch
from diffusers import DiffusionPipeline, UNet2DConditionModel, LCMScheduler
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file
base_model_id = "stabilityai/stable-diffusion-xl-base-1.0"
repo_name = "tianweiy/DMD2"
ckpt_name = "dmd2_sdxl_4step_unet_fp16.bin"
# Load model.
unet = UNet2DConditionModel.from_config(base_model_id, subfolder="unet").to("cuda", torch.float16)
unet.load_state_dict(torch.load(hf_hub_download(repo_name, ckpt_name), map_location="cuda"))
pipe = DiffusionPipeline.from_pretrained(base_model_id, unet=unet, torch_dtype=torch.float16, variant="fp16").to("cuda")
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
prompt="a photo of a cat"
# LCMScheduler's default timesteps are different from the one we used for training
image=pipe(prompt=prompt, num_inference_steps=4, guidance_scale=0, timesteps=[999, 749, 499, 249]).images[0]
1-step generation
import torch
from diffusers import DiffusionPipeline, UNet2DConditionModel, LCMScheduler
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file
base_model_id = "stabilityai/stable-diffusion-xl-base-1.0"
repo_name = "tianweiy/DMD2"
ckpt_name = "dmd2_sdxl_1step_unet_fp16.bin"
# Load model.
unet = UNet2DConditionModel.from_config(base_model_id, subfolder="unet").to("cuda", torch.float16)
unet.load_state_dict(torch.load(hf_hub_download(repo_name, ckpt_name), map_location="cuda"))
pipe = DiffusionPipeline.from_pretrained(base_model_id, unet=unet, torch_dtype=torch.float16, variant="fp16").to("cuda")
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
prompt="a photo of a cat"
image=pipe(prompt=prompt, num_inference_steps=1, guidance_scale=0, timesteps=[399]).images[0]
For more information, please refer to the code repository
License
Improved Distribution Matching Distillation is released under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
Citation
If you find DMD2 useful or relevant to your research, please kindly cite our papers:
@article{yin2024improved,
title={Improved Distribution Matching Distillation for Fast Image Synthesis},
author={Yin, Tianwei and Gharbi, Micha{\"e}l and Park, Taesung and Zhang, Richard and Shechtman, Eli and Durand, Fredo and Freeman, William T},
journal={arXiv:2405.14867},
year={2024}
}
@inproceedings{yin2024onestep,
title={One-step Diffusion with Distribution Matching Distillation},
author={Yin, Tianwei and Gharbi, Micha{\"e}l and Zhang, Richard and Shechtman, Eli and Durand, Fr{\'e}do and Freeman, William T and Park, Taesung},
booktitle={CVPR},
year={2024}
}
Acknowledgments
This work was done while Tianwei Yin was a full-time student at MIT. It was developed based on our reimplementation of the original DMD paper. This work was supported by the National Science Foundation under Cooperative Agreement PHY-2019786 (The NSF AI Institute for Artificial Intelligence and Fundamental Interactions, http://iaifi.org/), by NSF Grant 2105819, by NSF CISE award 1955864, and by funding from Google, GIST, Amazon, and Quanta Computer.