Feature request - HiDiffusion

[atxoxx]

Hi, could you add support for Hidiffusion ? It's a real game changer for sd1.5 and sdxl image generation basically it allows to make "native" sd1.5 image generation in 10241024,20482048 and for sdxl 20482048 and 40964096 without any issue (multiple people, strange limbs etc)

here's the github with the code : https://github.com/megvii-research/HiDiffusion

[yolain]

It's too hard for me.😂but i see the author is trying to implement in comfyui.

[atxoxx]

maybe I can help ? :D

[whmc76]

i have submit a request to the comfy hidiffusion node, maybe they will make a easier apply method

[yolain]

i have submit a request to the comfy hidiffusion node, maybe they will make a easier apply method

There are two custom nodes package so far, I did try it a couple of days ago but it didn't work very well.

https://github.com/blepping/comfyui_jankhidiffusion https://github.com/florestefano1975/ComfyUI-HiDiffusion

[atxoxx]

also this one : https://github.com/kijai/ComfyUI-ELLA-wrapper/commit/457c9b8a1ea5d98ba9c840481a7bf5552fdc7f20

maybe worth a look I mostly use this one and it work perfectly

[yolain]

also this one : kijai/ComfyUI-ELLA-wrapper@457c9b8

maybe worth a look I mostly use this one and it work perfectly

Yeah, I know this one too. But they all need to go through diffusers, not really native comfy.

[atxoxx]

So I guess best solution is to wait hidiffusion team to make something easier to implement