The latest version workflow and nodes are here
https://github.com/peterkickasspeter-civit/ComfyUI-ZImageTurboProgressiveLockedUpscale/tree/experiment
=====
Custom node: https://github.com/peterkickasspeter-civit/ComfyUI-ZImageTurboProgressiveLockedUpscale
Workflow is in the showcase images. More details here - https://www.reddit.com/r/StableDiffusion/comments/1r1qyj2/zimageturboprogressivelockedupscale_works_with_z/
Distill lora for base - https://huggingface.co/alibaba-pai/Z-Image-Fun-Lora-Distill/tree/main
*use either Z-Image-Fun-Lora-Distill-8-Steps.safetensors or Z-Image-Fun-Lora-Distill-4-Steps-2602-ComfyUI.safetensors
Description
Minor upgrades from v4
FAQ
Comments (18)
v5 example look to be re-introducing a lot more of the Flux-Face sharpness again.
The bad quality man himself 😅
Is that a good thing or a bad thing? I think it's because of the skin lora that I merged with v5. Any fixes or improvements you suggest?
@peterkickasspeter haha. your work has far surpassed mine at this point. Well done. I would just not merge that skin Lora as you mentioned. v4 was really good. How many training images are you working with right now?
@I_dont_want_karma_ i understand what you mean now. hmmmm. the skin texture lora I trained have ~50 images with mostly background blur. If i combine all the 3 lora's I used for the merge, the image count is around 200 (100 amateur photos, 50 shot on iPhone photos and 50 skin texture photos)
Just jumping in to say you two (peter and karma) are being wholesome.
After some testing of V5 I recommend a lower strength (0.3) compared to V4 (0.35-0.4). This compensates for the over-sharpness of faces while maintaining the better concept adherence of V5 compared to V4.
I find V5 to be an improvement, however slight compared to the significant jump seen in V4.
On the right track! Keep up the great work. 💪👍
Guys, what does "soap" mean? Even ChatGPT didn't know what it meant.
It means Shot On A Phone. It's in the model description
for instagram stories/posts, what resolution to use?
1440x1800
Trying to train a character lora for thiss, what do you reccomend? I tried to generate a base image then use pulid to generate the training photos but the lora combined with this made the images blurry, works well when i train a lora on real images and combine with this though. Anyone have any ideas?
did you get good results by training a character lora with real images and combining it with this one?
@aileverager555 Yeah, images not as good as with only this lora, but pretty good
@topgai I'm not able to understand. Blurry picture is related to sampler/scheduler
there are better amateur photo loras, try them. this one has some faults
@Latitude What's the best?
@Latitude leemme know
Latitude which ones?
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.