README.md

---
license: apache-2.0
library_name: diffusers
pipeline_tag: text-to-image
base_model:
- stabilityai/stable-diffusion-3.5-large
---
# Scale-wise Distillation SD3.5 Large
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](https://github.com/huggingface/diffusers) and [🧨 peft](https://github.com/huggingface/peft)
```
pip install -U diffusers
pip install -U peft
```

and then you can run 
<br>
```py
import torch
from diffusers import StableDiffusion3Pipeline
from peft import PeftModel

pipe = StableDiffusion3Pipeline.from_pretrained("stabilityai/stable-diffusion-3.5-large",
                                                torch_dtype=torch.float16,
                                                custom_pipeline="quickjkee/swd_pipeline").to("cuda")
lora_path = "yresearch/swd-large-4-steps"
pipe.transformer = PeftModel.from_pretrained(
    pipe.transformer,
    lora_path,
)

prompt = "Cute winter dragon baby, kawaii, Pixar, ultra detailed, glacial background, extremely realistic."
sigmas = [1.0000, 0.8956, 0.7363, 0.6007, 0.0000]
scales = [64, 80, 96, 128]

image = pipe(
    prompt,
    sigmas=sigmas,
    timesteps=torch.tensor(sigmas[:-1], device="cuda") * 1000,
    scales=scales,
    guidance_scale=1.0,
    height=int(scales[0] * 8),
    width=int(scales[0] * 8),
    max_sequence_length=512,
).images[0]
```
<p align="center">
<img src="large.jpg" width="512px"/>
</p>

## Citation

```bibtex
@inproceedings{
    starodubcev2026scalewise,
    title={Scale-wise Distillation of Diffusion Models},
    author={Nikita Starodubcev and Ilya Drobyshevskiy and Denis Kuznedelev and Artem Babenko and Dmitry Baranchuk},
    booktitle={The Fourteenth International Conference on Learning Representations},
    year={2026},
    url={https://openreview.net/forum?id=Z06LNjqU1g}
}
```