>>105741342
>Q5
go for Q8 anon, and if you don't have enough memory, offload a bit of that model to the RAM, like this, you don't want to miss that quality
https://github.com/pollockjj/ComfyUI-MultiGPU