Updated pics with locally trained lora "VETE" on 30/03/2026
Updated pics on illustrious XL2.0 05/03/2026
Mimic style of "POP929"
"V1.X" dataset is weak for eyes generation, use Facedetailer
"V1.Y" dataset expands x2.5, higher definition and better tags, can do close up portrait without facedetailer
No trigger words, but depending your base model; Tags such as 1990s_\(style\) or flat_colors can greatly help to make the styling lora to "pop out"
About V1.Y check models details ---->
Illustrious XL 2.0 V1.Y is trained on classic pattern 1024P / kohya / Official illustrious XL 2.0
V2 ILL V1Y is trained using beta version of civit AI toolkit at 1536P with civitai illustrious base
VETE is the locally trained lora, using VETE training checkpoint ; I believe this one is the closest to the artist styling. Start with 0.8 weight
Give both a try to see what does work the best for you, Illustrious XL 2.0 might be more flexible.
About V1.X:
V1 use custom illustrious model for training
V2 use baseline illustrious (civit) model for training
V3 is SD 1.5 Variant because why not ¯_(ツ)_/¯ , weight can be pushed up to 1.2 before it goes bad
base model used for generation is https://civarchive.com/models/76489?modelVersionId=111497 V2 yamime
and mostly Plantmilk "Walnut" for illustrious showcase pictures
Description
1024P trained on the last publicly available illustrious ( https://civitai.com/models/1369089/illustrious-xl-20 ) kohya/prodigy
I don't think that it's is better, simply different. Maybe more flexible.
FAQ
Comments (1)
My next public lora will probably cover 1888 | 1.123 :)