18/02/2026 - Bonus SDXL Lora added.
A pair of Fantasy Art Loras trained on the same 150 hand-picked Midjourney images.
Z-image doesn't really need help achieving the style and it just changes the composition.
The ZIT lora polishes the details in the refiner stage, it is a bit strong and I tend to use it around below 0.5 strength if generating with just ZIT.
I really recommended to use both LoRAs (ZIB and ZIT) in my 2-stage workflow: https://civarchive.com/models/2365846/jibs-double-turbo-zib-to-zit-workflow
The Idea is that is that ZIT is not as good at art Loras on its own
ZIB is better at art styles and prompt following, but lacks detail, so hopefully combining the 2 in a 2 Stage 2 model workflow gives the best of both worlds
The ZIB Lora does seem to be making Z-Image outputs a lot darker in this first version.
Description
Train Against my Jib MIX SDXL model but may work on others.
FAQ
Comments (3)
In your comparisons, you have Z-Image, Z-Image + Z-Image Fantasy Lora, and Z-Image + Z-Image Fantasy Lora + ZIT Lora. Is the "ZIT Lora" the ZIT version of this lora? Or is that some sort of Turbo lora for Base? I can't tell from the comparisons what's Base checkpoint, what's the Base version of the lora, and what's the ZIT version of the lora.
Its a 2 checkpoint workflow, so it uses Z-image with the lora for the first pass and then switches to ZIT with the ZIT version of the lora for a 2nd pass.
The First Image is the Base Z-Image Checkpoint.
The 2nd image is the same seed with the Z-Image Lora.
The 3rd image is the 2nd pass of image 2, with ZIT and the ZIT Lora.
The 4th is another single pass generation with just ZIT with the ZIT Lora.
Hope that makes sense.
After the second day of testing, I can say that this model is going into my favorites folder, great results on its own and in combination with other models, really great model...Thanx J1B for sharing!