Originally posted to HuggingFace by nerijs and shared on Reddit
Stable Diffusion v2-768 model trained on to generate isometric pixel art
How to Use
- Download the model and use it on your desired UI (Tested on AUTOMATIC1111's) Currently only .ckpt version is supported
- Trigger the style in your prompt with the isopixel token, look at the next section for more examples
Versions
- v1: 2500, 4000 and 5000 steps checkpoints available to download
Examples
isometric bedroom, isopixel style Steps: 50, Sampler: Euler a, CFG scale: 7.5, Size: 768x768
isometric sushi store, isopixel style Steps: 50, Sampler: Euler a, CFG scale: 7.5, Size: 768x768
isometric gas station, isopixel style Steps: 50, Sampler: Euler a, CFG scale: 7.5, Size: 768x768
isometric magical forest, isopixel style Steps: 50, Sampler: Euler a, CFG scale: 7.5, Size: 768x768
Tips
- Always use 768x768
- High step count on Euler_a gives the best results
- Low CFG scale outputs great results
- You can use a tool like Pixelator to achieve a better effect. This model isn't pixel perfect (yet 😉)
Please consider supporting further research on my Patreon
If you have any question, suggestion for new models or need help in general with SD related stuff, don't hesitate to reach out on Twitter
Description
FAQ
Comments (3)
You'll need to take the v2-inference-v.yaml file from stable diffusion and put it alongside the model in the same directory. You can right click on "raw" just above the file and Save link as to download the file > then rename it so that the filename is the same as the checkpoint file - isopixelDiffusion_v15000.yaml > then restart webui
ERROR doesn't work. Missing YAML File? Corrupted? Not sure.