[Feature Request]: Hotswap modelscope models to support an easier workflow with zeroscope and the upscaling model

[itspoidaman]

Is there an existing issue for this?

• [X] I have searched the existing issues and checked the recent builds/commits

What would your feature do ?

Zeroscope V2 dropped, a model trained ontop of modelscope, but it comes in two parts. A model that you use for base generation, a model you use for upscaling. There's also the Potat1 model

Requesting a feature that allows you to swap between the different modelscope custom models without having to rename/replace the file that it automatically points too, would also be nice if this feature dumped the currently loaded model from the memory pipeline

Proposed workflow

1. select which model you want to load with a drop down menu
2. it loads the model for you

Additional information

No response

[Yaiol]

I found a way to make things a bit easier, without needing to rename/replace the files Simply prepare two configuration files Change the model_args variables ckpt_clip, ckpt_unet, ckpt_autoencoder

    "model_args": {
        "ckpt_clip": "zeroscope-v2-576x320-clip.bin",
        "ckpt_unet": "zeroscope-v2-576x320-unet.pth",
        "ckpt_autoencoder": "kabachucha-autoencoder.pth",
        "max_frames": 16,
        "tiny_gpu": 1
    },

And then create as much batch you need copy zeroscope-v2-576x320.json configuration.json