Hi, I’m very new to all of the AI stuff and even my Python is a little rusty.
I got some of the text-to-image examples working on an Anaconda installation with Jupyter under VSCode on Windows, but the text-to-image example in the docs using runwayml/stable-diffusion-v1-5 takes about 5 minutes for an image (cuda is working, the task manager shows the GPU being used).
My first goal would be to have some base, potentially with a much simpler model, that can create an image in about 10s so that I can experiment a bit better. The model doesn’t have to be useful, it would be fine if it can create only certain peculiar images in a low resolution.
I’ve looked around the model hub, but there doesn’t seem to be a way to find a particularly small one.
I’m also having some technical issues that may contribute to the slowness.
I’m usually getting a message like
safety_checker\model.fp16.safetensors not found
or
vae/diffusion_pytorch_model.fp16.safetensors not found
Those are files that I’ve checked are in the respective GitHub repo. Maybe that means that fp16 isn’t used, but I don’t know how to check that.
I also tried using num_inference_steps set to a very small value, but that always gives me a black square telling be that’s it’s likely NSFW…