Model Selection

[mosquitobird11]

On September 7 (in two days from now) Haru is going to release a finetuned SD1.4 model for anime-like generations, trained on a set of 56,000 high aesthetic CLIP-filtered Danbooru images.

This, along with the suspected 1.5 SD coming in the near future, would be more accessible if we had a way to easily switch between models in the webUI, even though this would require releasing and reloading the new model into memory.

[Muskelmagier]

Yeah, with another model i follow being released in the near future, this becomes even higher priority

[lendrick]

The waifu diffusion model can be downloaded here if anyone is working on this and needs another model to test with:

https://storage.googleapis.com/ws-store2/wd-v1-2-full-ema.ckpt

[mosquitobird11]

@Muskelmagier @lendrick and anyone else who wants a quick fix for this, this is what I did for now:

in webui.py, at the beginning of load_model_from_config I added the lines:

    all_models = {
        "1": "models/ldm/stable-diffusion-v1/model.ckpt",
        "2": "models/ldm/stable-diffusion-v1/model_waifu1.2.ckpt",
    }

    while 1:
        modelnum = input('What model number?\nEnter 1 for: Stable Diffusion 1.4\nEnter 2 for: Waifu Diffusion 1.2\n')
        if modelnum in all_models:
            ckpt = all_models[modelnum]
            break

Obviously you have to have models in that location named like that for it to make sense, but this is pretty extensible and easy for me to choose a model on boot without renaming models etc.

[Batman313v]

So if this is the before:

You pasted it here?

[mosquitobird11]

Yeah that's correct! The code is super hacky / simplistic but it should be easy to follow :)