I set the base version early access to recoup my runpod costs lmao
Workflow: https://pastebin.com/nvc0f8Qc
Custom node: https://github.com/peterkickasspeter-civit/ComfyUI-ZImageTurboProgressiveLockedUpscale
Description
FAQ
Comments (11)
This LoRA came out really well. Do you have any tips to increase size further, or to make sure we get high skin detail? Workflow seems to go smoothly.
if you are not too picky about the composition changing a bit, you can always increase the tail first upscale and tail last upscale values to 6 and 4 (so it starts at 6 and then gradually ends at 4). you can also decrease the max scale factor to 1.40 and use an upscaler like ultrasharp so it runs a bit longer.
if you really want to push it, enter at a higher refine sigma (like 0.78) while using model sampling aura flow value of 5.
the training dataset has barely any faces so it tends to add lots of crazy details on the breasts but faces sometimes look like they are coming from the base model
Silly question but can it do good rear view shots I tried making my own character Lora and she had a large chest but because of a lack of rear view photos it didn't turn out well. Any recommendations on how many images I should put for the training for rear view photos with large chest. Only anime models tend to well at this 😂
rare view is what? like the subject is turned back? someone else asked me this question before but i dont understand it well. if you have an image or prompt i can test and see if this lora can do it or not
@peterkickasspeter
Not OP, but I think that they want something that would trigger being labelled with "backboob" and "from behind".
Here is a simple image on this site that does those things. But without sagging breasts, and using tags.
https://civitai.com/images/25400078
@sawbladex818 ahh like that. okay i just posed https://civitai.com/images/121570904
@peterkickasspeter that's super That's exactly what I wanted. It's only because I created a character Lora and the rear view was horrible 😂😂😂😂😂 like it put boobas on the back 😂😂 Great work as always good sir.
Do I need sage attention with this workflow? I don't use it but it's not giving me the best results
you can bypass it. its not mandatory.
Is the Zbase version like a version 2.0 or Zturbo or does it work about the same.
With your updated ZImageTurboProgressiveLockedUpscaleVersion1 node what settings do you recommend also wil it work with the new Z-Image-Fun-Lora-Distill-4-Steps-2603-ComfyUI Lora? Or the 8 step variant.