Overview
This is the Z-Image-Turbo version of my VestalWater's Illustrious Styles for Qwen Image lora. This LoRA aims to make ZIT's output look more like images from an Illustrious finetune.
If you like my work, please consider supporting me at https://ko-fi.com/vestalwater.
Why use Z-Image-Turbo + an Illustrious Lora over Illustrious alone? ZIT has great prompt adherence and handles complex prompts really well, but it doesn't render images with the most flattering art style. Illustrious is the opposite: It has a great art style and aesthetics but struggles as soon as the prompt demands complex subject positions and specific elements to be present in the composition.
This lora aims to capture the best of both worlds, ZIT’s understanding of complex prompts and the lora adds a (subjectively speaking) flattering art style on top of it.
V2 is out!
VestalWater's V2 Lora is out! This new version features the following improvements:
Completely new dataset using extremely cherry picked illustrious images. This results in a far more coherent background and significantly less hallucinatory and messy lines in the background.
The new dataset also removes the weird striations on women's skin.
Used version 2 of Ostris' Training Adapter which should make the model understand the concept better
Comparison between v1 and v2
Same prompt, not cherry picked, keep in mind the turbo version of ZIT has a problem with hands. High res comparison link: https://civarchive.com/posts/24879507
V2
V1
Settings and Workflow
All images in the image carousel have ComfyUI workflows attached.
TLDR:
Sampler: ER_SDE (Please don't use euler simple. It's so bad for illustrations.)
Scheduler: Simple
Lora Strength: 0.8
Steps: 8
CFG: 1
Description
FAQ
Comments (7)
Any idea how to get zImage + this lora to generate a photo realistic image rather than a drawn image. When I use your lora I always get a drawn image, despite positive and negative prompts that should make it photo realistic.
I would mix my lora with a realistic instagram girl lora at a 0.3:0.7 strength ratio. Keep messing with the strength ratio and the lora combination until you get the look you want.
@VestalWater Thanks, I will give it a try.
Was using comfyui v0.3.76 then updated to latest version, and somehow it has way worse results with same parameters :(
I recently updated from 0.3.75 to 0.6.0 and my results are coming out different as well. I think they've done something that makes the influence of loras to be a lot stronger. So if you were using 0.8 lora strength with version 0.3.75, then with the latest version you should use about 25% less or 0.6. Look at this comparison to see what I'm talking about: https://civitai.com/images/115723635
@VestalWater oohh i see, interesting, ty for info
lora is very good but need one for klein 9b for image edit.