Using fp32 VAE & fp16 VAE in the same workflow without restarting comfyui
- Dominant language
- Python
- Stars
- 133k
- Forks
- 15.7k
- Avg merge
- 1d 10h
- Merged PRs (30d)
- 153
Description
Hi, let me introduce :
I have a gtx 1660 ti 6gb , and when vae decoding an upscaled img, comfyui sometimes switch to tiled vae and i don't like that (ugly color, less details). So, to counter that i use the --fp16-vae command line and no more tiled vae is needed (work 95% of the time, 5% are black img but it's ok). The problem is fp16 vae on my gpu take time.. is absolutly ok when is for decoding upscale, but for regular (i use mainly sd 1.5 so 512*768) it's becoming annoying. Ok it's only few seconds more (with lcm i take 3x more time to vae decoding than to generate).
So i was thinking : Is there a way or a node to select fp32/fp16 when decoding without restarting comfyui ?
For ex: be able to fp32 vae decoding the 512*768, but fp16 vae decoding the x2 upscaled version ?
I have no idea if it's possible, or if a node exist already for that, so thanks in advance (and hope i explained correctly)
Contributor guide
Assessment
This issue has not been assessed yet.