---
license: apache-2.0
library_name: diffusers
pipeline_tag: text-to-image
---
# Scale-wise Distillation 3.5 Medium
Scale-wise Distillation (SwD) is a novel framework for accelerating diffusion models (DMs) 
by progressively increasing spatial resolution during the generation process. 
<br>SwD achieves significant speedups (2.5× to 10×) compared to full-resolution models 
while maintaining or even improving image quality.
![3.5 Medium Demo Image](swd.png)

Project page: https://yandex-research.github.io/swd <br>
GitHub: https://github.com/yandex-research/swd <br>
Demo: https://huggingface.co/spaces/dbaranchuk/Scale-wise-Distillation

## Usage
Upgrade to the latest version of the [🧨 diffusers library](https://github.com/huggingface/diffusers)
```
pip install -U diffusers
```

and then you can run
<br> (Probably, you will need to specify the visible device: %env CUDA_VISIBLE_DEVICES=0, for correct loading of LoRAs.)
```py
import torch
from diffusers import StableDiffusion3Pipeline
from peft import PeftModel
pipe = StableDiffusion3Pipeline.from_pretrained("stabilityai/stable-diffusion-3.5-medium",
                                                torch_dtype=torch.float16,
                                                custom_pipeline='quickjkee/swd_pipeline')
pipe = pipe.to("cuda")
lora_path = 'yresearch/swd-medium-4-steps'
pipe.transformer = PeftModel.from_pretrained(
    pipe.transformer,
    lora_path,
)
generator = torch.Generator().manual_seed(1)
prompt = 'A dog holding a sign that reads Sample Faster'
sigmas = [1.0000, 0.9454, 0.7904, 0.6022, 0.0000]
scales = [32, 64, 96, 128]

images = pipe(
    prompt,
    sigmas=torch.tensor(sigmas).to('cuda'),
    timesteps=torch.tensor(sigmas[:-1]).to('cuda') * 1000,
    scales=scales,
    guidance_scale=0.0,
    height=int(scales[0] * 8),
    width=int(scales[0] * 8),
    generator=generator,
).images
```
<p align="center">
<img src="medium.jpg" width="512px"/>
</p>

## Citation

```bibtex
@article{starodubcev2025swd,
  title={Scale-wise Distillation of Diffusion Models},
  author={Nikita Starodubcev and Denis Kuznedelev and Artem Babenko and Dmitry Baranchuk},
  journal={arXiv preprint arXiv:2503.16397},
  year={2025}
}
```