Idea, it'd be nice it we could add hugginface fp16 variants easily

[paper quiver]

As a user with limited shared ram, it take ages for Invoke load a 32bit model and convert it to 16bit.
I guess it would also be an issue with users with limited VRAM needing space for the conversion.
It would be nice for users to have an easy way for users like me to add the 16bit variants from the model configuration tools
without have to download then from hugginface and manually add them

Is the GIU model configuration tool refers to a different type of variant I add this as an example of what you'd do in diffusers

refiner = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-refiner-1.0", torch_dtype=torch.float16, variant="fp16").to('mps')

This would down load a fp16 version of the model if it exists in the repo.

[sour sail]

https://civitai.com/models/117188
So I just downloaded this one today (it's the base 16'er, prior DL'd claimed 16 but when came to be about it... Mmnmm.) And I noticed it didn't seem as powerful as the 32. I know I've heard it a million times over--that 32bit 1.5 models are pretty much dead-on-par with 16ers--that their main use is really if you plan to mix.

But I noticed my first bit of odd eyes or facial distortion with this one. It wasn't anything near the atrocious affects that would randomly roll from the dice cup in 1.5, given the prompt/model circumstances had aligned under the devil's &*%hole.