coralchar-diffusion / README.md
nerijs's picture
Update README.md
46d8050
---
license: creativeml-openrail-m
---
<div style="display: flex; flex-direction: row; flex-wrap: wrap">
<a href="https://www.patreon.com/user?u=29466374" target="_blank">
<img src="https://img.shields.io/badge/Patreon-F96854?style=for-the-badge&logo=patreon&logoColor=white" alt="Patreon"/>
</a>
<a href="https://twitter.com/nerijs" target="_blank">
<img src="https://img.shields.io/badge/Twitter-1DA1F2?style=for-the-badge&logo=twitter&logoColor=white" alt="Twitter"/>
</a>
</div>
# coralchar-diffusion-v1
Stable Diffusion v1.5 model trained on to generate cute character portraits
<div style="display: flex; flex-direction: row; flex-wrap: wrap">
<img src="https://s3.amazonaws.com/moonup/production/uploads/1670205150413-6303f37c3926de1f7ec42d3e.png" width="256">
<img src="https://s3.amazonaws.com/moonup/production/uploads/1670205171617-6303f37c3926de1f7ec42d3e.png" width="256">
</div>
## How to use
- Download the model and use it on your desired UI (Tested on AUTOMATIC1111's) .ckpt and Diffusers version available
- Trigger the style in your prompt with the **coralchar** token, look at the next section for more examples
- If you want to use the inpainting model, you can use it like a normal v1.5 model
## Versions
- **v1**: 1000-6000 steps checkpoints available to download
- **inpainting** version available
## Examples on step-6000 model
**a woman wearing blue jeans and a white tank top**
Steps: 20, Sampler: DPM++ SDE, CFG scale: 7, Size: 512x768
<img src="https://s3.amazonaws.com/moonup/production/uploads/1670204360798-6303f37c3926de1f7ec42d3e.png" width="512"/>
**a man wearing a black puffy vest**
Steps: 20, Sampler: DPM++ SDE, CFG scale: 7, Size: 512x768
<img src="https://s3.amazonaws.com/moonup/production/uploads/1670204467592-6303f37c3926de1f7ec42d3e.png" width="512"/>
## Examples on inpainting model
**a man wearing a blue puffy vest**
Steps: 20, Sampler: DPM++ SDE, CFG scale: 7, Size: 512x768, 0.75 Denoising strength
<h2>Original vs step_6000 vs inpainting version</h2>
<div style="display: flex; flex-direction: row; flex-wrap: wrap">
<img src="https://s3.amazonaws.com/moonup/production/uploads/1670205036420-6303f37c3926de1f7ec42d3e.png" width="256"/>
<img src="https://s3.amazonaws.com/moonup/production/uploads/1670204708270-6303f37c3926de1f7ec42d3e.png" width="256"/>
<img src="https://s3.amazonaws.com/moonup/production/uploads/1670204954426-6303f37c3926de1f7ec42d3e.png" width="256"/>
</div>
## Tips
- Best results with 512x768, outputs full body portraits
- Also high step count on Euler_a gives good results
- Low CFG scale outputs great results
- If you want to generate different expressions, generate a base character with txt2img then adjust your outfit and details with inpainting model and use inpainting again to generate different expressions and poses
Please consider supporting further research on my Patreon:
<a href="https://www.patreon.com/user?u=29466374" target="_blank">
<img src="https://img.shields.io/badge/Patreon-F96854?style=for-the-badge&logo=patreon&logoColor=white" alt="Patreon"/>
</a>
If you have any question, suggestion for new models or need help in general with SD related stuff, don't hesistate to reach out on Twitter:
<a href="https://twitter.com/nerijs" target="_blank">
<img src="https://img.shields.io/badge/Twitter-1DA1F2?style=for-the-badge&logo=twitter&logoColor=white" alt="Twitter"/>
</a>