I wanted to make a nix-approach to stable difussion models.
Some of the models are combinations of others, so basically if you have a declarative, hash-checked way to generate models based on others, you can have a subset of base models and all goes from there.
This way the exact combination of models can be kept, instead of it being lost to time and civitai
Also, within the same project, have stable-difussion generation nixified, so it is easy to replicate results.
I wanted to make a nix-approach to stable difussion models.
Some of the models are combinations of others, so basically if you have a declarative, hash-checked way to generate models based on others, you can have a subset of base models and all goes from there.
This way the exact combination of models can be kept, instead of it being lost to time and civitai
Also, within the same project, have stable-difussion generation nixified, so it is easy to replicate results.
If you are interested we could coordinate it :)