How to setup a offline folder for the safetensors that work?

[nnqq2603]

Provide Stable Diffusion 2.1 diffusers weights

1. The node defaults to stabilityai/stable-diffusion-2-1-base (512px).
2. For sharper results you can use the 768px repo stabilityai/stable-diffusion-2-1.
   Example offline download:
3. huggingface-cli download stabilityai/stable-diffusion-2-1 --local-dir "<your path>" --local-dir-use-symlinks False
4. Point the base_model_path input at the chosen folder or keep the repo id to download on demand.
   ==========

The #4 is the point I can't do. I already git clone the entire huggingface repo (https://huggingface.co/EricRollei/SAI-SD2.1-for-HYPIR) contains all necessary files (included the safetensors) into a folder in my local system.

*\comfy\ComfyUI\models\HYPIR\SAI-SD2.1-for-HYPIR\

And I enter *HYPIR\SAI-SD2.1-for-HYPIR* to base_model_path input field, but it doesn't work. How exactly to format the folder to the proper diffuser-format directories like the tooltip text seems to hint at?

[GeffersAtWork]

I answered this on some other page, maybe another repo, so not sure. But your folder structure looks like mine, except my path is F:\ComfyUI_windows_portable\ComfyUI\models\HYPIR\stable-diffusion-2-1-base (the first part is obviously down to your installation) I'll do a complete dir /s if this is still not right for you. I also don't have that safetensors file seemingly at all.

[nnqq2603]

I answered this on some other page, maybe another repo, so not sure. But your folder structure looks like mine, except my path is F:\ComfyUI_windows_portable\ComfyUI\models\HYPIR\stable-diffusion-2-1-base (the first part is obviously down to your installation) I'll do a complete dir /s if this is still not right for you. I also don't have that safetensors file seemingly at all.

Thanks, it did run and output result which is different from the input image, so I assume it ran well but the quality and realistic degree of the output result is very poor to be honest. I assume the code is demo or the entire repo is proof of concept more than intensive upscale model with real life practical usefulness.

[GeffersAtWork]

I've been using "Waifu2x Interpolater", which is free, and runs outside of Comfy. I'm sure you would find it with a google search. It has a great batch feature where you drag and drop videos, or images, and interpolates frames/and/or upscales. It seems a ton faster than any Comfy solution, though I have been mainly using the frame interpolator rather than the upscaler.