README.md

---
language:
- ja
- en
---
# yumekawa_diffusion  

[![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1jUvv2RXEFo-fQVSSzVBs7vU22au3OLcC?usp=sharing#scrollTo=Ri6jRc9kAust) 
Generate a "Yumekawaii" image  

Do not use too many Prompts.   

Negative Prompt should be low quality, bad face, bad hands, etc.  


<img src="https://s3.amazonaws.com/moonup/production/uploads/1673848269039-63056e2d99870e13d3df4e73.png" width="900" >

# How to run 

You can run in Google colab[![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1jUvv2RXEFo-fQVSSzVBs7vU22au3OLcC?usp=sharing#scrollTo=Ri6jRc9kAust) 

Use [🤗HuggingFace Diffusers library](https://github.com/huggingface/diffusers)  

clone the library  
```
pip install --upgrade git+https://github.com/huggingface/diffusers.git transformers accelerate scipy
```
and run folowing scripts 

```python
import torch
from diffusers import StableDiffusionPipeline, EulerDiscreteScheduler

model_id = "teftef/yumekawa_diffusion_ver2.1"

scheduler = EulerDiscreteScheduler.from_pretrained(model_id, subfolder="scheduler")
pipe = StableDiffusionPipeline.from_pretrained(model_id, scheduler=scheduler)
pipe = pipe.to("cuda")

prompt = "a girl"
negative_prompt="low quality,bad face,bad anatomy,bad hand"
image = pipe(prompt,negative_prompt=negative_prompt).images[0]

image.save("output_img.png")
```

# Result

**Left** : Waifu Diffusion 1.4 e1  
**Middle** : Fine tuned model without LoRA  :     **[can use in here]**(https://huggingface.co/teftef/teftef_fox_ear_2_0/tree/main)  
**Right** : Fine tuning with RoLA (yumekawa_diffusion)  

### Fixed Prompts
・Prompt : a girl,  
・Negative Prompt : low quality,bad face,bad hands,  
・Step : 28  
・CFG Scale : 7.5  
・Sampling method : Euler a  
・ｗ×ｈ＝512×768  



<img src="https://s3.amazonaws.com/moonup/production/uploads/1673772447629-63056e2d99870e13d3df4e73.png" width="700" >


### Various Prompts  
・Negative Prompt : low quality,bad face,bad hands,  
・Step : 28  
・CFG Scale : 7.5  
・Sampling method : Euler a  
・ｗ×ｈ＝512×768  


<img src="https://s3.amazonaws.com/moonup/production/uploads/1673771856440-63056e2d99870e13d3df4e73.png" width="700" >

Traning : 2023/01/13  
Public : 2023/01/15 teftef