Trigger phrase is: "fake ass", or "huge fake ass".
Anima: Dataset is IRL photos, recommend higher strength compared to other models. Effect is stronger without artist tags.
Ideogram-4: Lora seems a bit overfitted at 1.0 strength, recommend lower strength say 0.4-0.6.
Krea2: 1 strength plus or minus a few tenths recommended. Starts to burn images around 2 strength.
MOSS: Any strength up to 1.5 or so seems to work best.
Qwen 2.1: Make sure to increase the CFG Guidance to 3+.
Z-Image: 0.6-1 seems to be a good strength.
My loras are always free to download and generated at minimum possible cost. If you'd like, please consider a tip: buymeacoffee.com/auxilia
Description
Cleaned up my dataset, Trigger phrase is just "fake ass" for Ideogram, and the strength should influence the size.
FAQ
Comments (4)
Anima version maybe?
This dataset is all IRL photos, isn't anima for cartoon/anime?
@auxilia Some finetunes are good at realism to a certain point that mimic pony-realism back then, Reilgun the lora creator on this site already created some breast implants lora using 50% anime and 50% realistic dataset
@Pupper I'll see how it turns out after Ideogram4. I have to use layer offloading to make Loras so it's pretty slow. Good thing Anima isn't too heavy.