Gets rid of the blocky skin details of Z-Image on higher resolutions.
Trained on 300+ very highly detailed high resolution (min 2000px) images of amateur female face portraits, with visible skin pores, skin hairs and blemishes/moles.
Use on LoRa strength around 0.7
Higher strength will change the face/identity -> it will look more western/latina . (If you are looking for blazing high contrast quality more face-oriented generation you may try to more agressively trained Female Tongue, Mouth and Teeth)
Gallery side-by-side images (Euler, default ComfyUI workflow , 1536x1536px)
* Left is Z-Image default - no LoRa
* Right is Z-Image + this LoRa on 0.7
Any feedback appreciated for next versions!
Description
FAQ
Comments (28)
I didn't try the Lora yet but the left side just looks like you're using the wrong sampler.
for some reason i prefer the left side more
Euler
@fox23vang226 Curious what is it about the left that looks better than the right? Also this lora is trained on primarily caucasian and latina women, so you can also use it to bias the face away from the std faces in the model to get some variety on higher strenghts , but it's really up to you if its useful to you or not of course!
i hate loathe the left side.
Yeah, I'm using euler_ancestral / beta57 and not getting this result at all. Skin is clear here.
@jhnprst2150 I guess Im just sick of the overly airbrushed and perfectly smooth AI face look. I like the rough look.
@fox23vang226 gotcha, but if i am right the lora does definately not introduce airbrush/smooth skin.. my photos in the dataset are all far from perfect, all amateur faces raw photography with lots of skin imperfections, pores etc.
@PlayAI "hate"? just against an image?
@BIG_A I fixed it
Which side is the Lora? @jhnprst2150
Gallery side-by-side images (Euler, using default ComfyUI workflow )
* Left is Z-Image default - no LoRa
* Right is Z-Image + this LoRa on 0.7
You're doing something wrong with the close-ups; I just did it without the lora and they look perfect.
allright, which sampler and steps did you use?
if your results without lora are better than with i am curious to compare myself!
i am using default euler or dpm++2m sampler and gen on 1024 , 1280 or 1536 px
thanks!
Thank you for this wonderful LoRa. Your quality work gives me FLUX level quality Close-Ups. Pls also for HunyuanVideo1.5, like the good old times...
thank you, its just the first version; i am trying to see if can improve, and also eagerly awaiting the base model release ;-)
ahah hunyanvid I have not tried it yet, did you? is it any better than WAN? I am using 2.1/2.2 right now for local vid and cant imagine it can get better (I love WAN 2.1 VACE)
@jhnprst2150 I tried it, and this workflow embedded in this video gives some really nice quality and fast results: https://civitai.com/images/111425198
not the official workflow, but somehow his approach is really good
@TedBerlim thanks will check it, i read a lot of comments that it does not really surpass WAN , and even has slower rendering and occasionally worse movements, so it did not really seem worth the trouble; but it cannot hurt to try again one time, thanks
Nice!
Hey, this would fit better into concept category. The poses category is mainly for full body poses, while concept category is for a broader range of loras.
allright, changed it
ok, thank you
An update would be needed because unfortunately by updating ComfyUI this LoRa no longer works well and it's a shame because it was one of those that I had put in the favorites in the LoRa Manager
hey? what kind of error you get?
使用了这个LORA，但是好像没有什么作用。必须是脸部特写才能有效果吗？
it was trained on faces mosty yes, so closeups or at least images with a lot of skin you would see the effects best
Excellent work! This is somehow fixing my inconsistent Character Loras.
Details
Files
FemaleFacePortraitsDetailedSkin-ZImage.safetensors
Mirrors
189_FemaleFacePortraitsDetailedSkin-ZImage_dec18.safetensors
FemaleFacePortraitsDetailedSkin-ZImage.safetensors
FemaleFacePortraitsDetailedSkin-ZImage.safetensors
FemaleFacePortraitsDetailedSkin-ZImage.safetensors
FemaleFacePortraitsDetailedSkin-ZImage.safetensors
FemaleFacePortraitsDetailedSkin-ZImage.safetensors
FemaleFacePortraitsDetailedSkin-ZImage.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.