README.md

---
datasets:
- huggan/few-shot-aurora
---

<center>

![Aurora](https://huggingface.co/li-yan/diffusion-aurora-256/resolve/main/doc/Aurora.gif)

![Aurora Photo](https://huggingface.co/li-yan/diffusion-aurora-256/resolve/main/doc/Aurora-by-Li-Yan.jpg)

</center>

# Description

Have you ever seen aurora with your own eyes? Check the above picture I got in Alaska in Winter. Beautiful right?

However, aurora is so rare that we can hardly see it even in the very north places like Alaska.

Don't worry. Now we have generative models!!! Here are the pictures generated by this model:



| ![sample1](https://huggingface.co/li-yan/diffusion-aurora-256/resolve/main/doc/sample_1.png) | ![sample1](https://huggingface.co/li-yan/diffusion-aurora-256/resolve/main/doc/sample_2.png) | ![sample1](https://huggingface.co/li-yan/diffusion-aurora-256/resolve/main/doc/sample_3.png) | ![sample1](https://huggingface.co/li-yan/diffusion-aurora-256/resolve/main/doc/sample_4.png) |
|--|--|--|--|
| ![sample1](https://huggingface.co/li-yan/diffusion-aurora-256/resolve/main/doc/sample_5.png) | ![sample1](https://huggingface.co/li-yan/diffusion-aurora-256/resolve/main/doc/sample_6.png) | ![sample1](https://huggingface.co/li-yan/diffusion-aurora-256/resolve/main/doc/sample_7.png) | ![sample1](https://huggingface.co/li-yan/diffusion-aurora-256/resolve/main/doc/sample_8.png) |



# Model Details

This model generate 256 * 256 pixel pictures of aurora.

It is trained from dataset [huggan/few-shot-aurora](https://huggingface.co/datasets/huggan/few-shot-aurora).

The training method is modified from this [example](https://colab.sandbox.google.com/github/huggingface/notebooks/blob/main/diffusers/training_example.ipynb).

You can check my training source code here: [<img src="https://colab.research.google.com/assets/colab-badge.svg">](https://colab.sandbox.google.com/github/Li-Yan/Diffusion-Model/blob/main/li_yan_diffusers_training_accelerate.ipynb)

# Usage

## Option 1 (Slow)
```python
from diffusers import DDPMPipeline

pipeline = DDPMPipeline.from_pretrained('li-yan/diffusion-aurora-256')
image = pipeline().images[0]
image
```

## Option 2 (Fast)
```python
from diffusers import DiffusionPipeline, DDIMScheduler

scheduler = DDIMScheduler.from_pretrained('li-yan/diffusion-aurora-256')
scheduler.set_timesteps(num_inference_steps=40)

pipeline = DiffusionPipeline.from_pretrained(
    'li-yan/diffusion-aurora-256', scheduler=scheduler)

images = pipeline(num_inference_steps=40).images
images[0]
```