ComfyUI Workflow
visit this guy hes doin some awesome images, models and workflows too :)
the images in the preview are all created with SDXL (just saying)
After playing a while with his workflow i did the usual things and added the stuff i have in my other workflows. (with his permission xd) but don't worry there are no fancy nodes in here^^
there are 2 more detailers (B & V) (you know why)
added sd3 support
you can choose either to use the standard model or the one with the included text encoder. the tripple clip loader can be bypassed.
added a lot of options to enable/bypass things
it should be obvious but i'll say it anyway: only one of the two groups should be activated, either sd15/sdxl or sd3
i usually don't build single image workflows but this was kind of fun to do... Well it's a rebuild not a build from scratch.
Description
some fixes are done
ReUploaded because i forgot to connect the lora stacker, sorry for that.
the aspect ratio node caused some issues with the renders, notable in sd3 (reported by jadewind)
while comparing the output with a standard workflow i noticed that i wasn't able to create the same result with this workflow. soooooo...
if the clip select last layer is active with an sdxl checkpoint the output is completly different. haha i know its intended but it can't be disabled. so the efficient loader is gone, the standard nodes are back. (the image is more important than convenience xd)
the clip select last layer can be disable and it is by default. (set to 1)
the vae input switch is by default=1, means baked vae is used.
i added perturbed attention guidance and self attention guidance. result is nice. you can use them all together or one of them but i would not recomment to use them together with kohya deep shrink.
this version is more or less a fix. actually i wanted to update more but fixing the bugs was more important for now.