Type | |
Stats | 930 |
Reviews | (53) |
Published | Feb 20, 2023 |
Base Model | |
Training | Epochs: 20 |
Trigger Words | citrinne |
Hash | AutoV2 7F4305ACFD |
Update: Added AOM3 version.
Back with another FE Engage model. This time for Citrinne!
There is not a lot of art out there yet for her so the dataset was pretty limited on this one. That being said I think it turned out fairly well minus a few quirks.
Quirk #1: I've noticed the LORA likes to throw borders on images, I'd suggest adding negative prompts (like "border") if you want to avoid those.
Quirk #2: LORA does a good job with her jewelry overall, but doesn't have enough data to completely understand the asymmetrical earrings consistently yet. Hoping I can train a LORA with more references down the line.
Based on Anythingv3, but appears usable on many anime-focused models. Trained on Clip Skip 2, but seems capable of being used on 1 as well.
I've seen good results at a lot of different strengths, but 0.7+ is usually needed to get the jewelry consistently. The higher the strength, the more outfits tend to conform to her canon one. Throwing prompts like "dress" and "detached sleeves" can help as well.
For more specifics on my normal prompting methods check out the description on my Yunaka model here!