Midjourney LoRA!
Note: the Z-Image Turbo LoRA is compatible with BOTH models. Z-Image Base and Turbo.
HIDREAM LORA DOES NOT WORK WITH 2604.
BASE ONLY WORKS WITH BASE ONLY!
base only is recommended to run at 2048 X 2048
BASE VERSION HAS BEEN UPDATED TO A FURTHER TRAINED CHECKPOINT!
(I reuploaded a new version of the Base lora. my previous version was an unfinished early checkpoint save because my session was corrupted. i have since retrained the lora and it is now the full training session! redownload if you already did!)
RECOMMENDED STRENGTH:
0.2 - 0.8
adds the midjourney aesthetic to your generations. VERY well trained to be as close to Midjourney as possible.
Dataset: 1217 Images, various styles, angles, lighting, aesthetics, contrasting, subjects, etc. the dataset is DENSE.
WARNING: LORA IS STRONG. i recommend testing on 0.1 and working your way up to the desired strength per image.
Description
pruned dataset compared to previous iterations. sadly civitai limits dataset count. original dataset was 1500 images, this version is only 1000 so not too much of a trade off but it is missing some key aesthetics i had to remove.