Trying to use GFPGAN gives an Runtime Error - (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same about stable-diffusion HOT 4 CLOSED

Well why are you using --gfpgan-cpu --esrgan-cpu? you don't need to
esrgan works on cpu, it seems gfpgan doesn't

from stable-diffusion.

I was running GFPGAN on CPU because running it on GPU caused OOM errors, oh well, if it doesn't work on CPU then not much can be done. Thanks anyways.

from stable-diffusion.

You could try using gfpgan on it's standalone tab as post-processing, and also try removing the model file for esrgan so that doesn't load, I suppose I should add an option to disable loading each of the extra models all together but you can do the same thing for now by just deleting (or moving) the appropriate .pth model file

6gb is towards the lower end of the absolute bare minimum requirements though so issues with OOM are to be expected really.

from stable-diffusion.

Got it, that seems like a good idea. Thank you for your help!

from stable-diffusion.