Update: Added an AOM3 version.
Hiya Papaya! This is my first attempt at putting together a LORA.
Yunaka is a bit of a challenge due to the stars on her face and in her hair. I've noticed strengths 0.7-0.8 are good for catching her likeness with limited success on the stars. 0.9+ seems to do a good job at both with the tradeoff of more difficulty changing hair/outfits.
Would love to get some feedback on how to improve!
Lora is trained on Anythingv3, but seems to work well with a variety of anime-focused models like AbyssOrange and Grapefruit.
My usual settings:
Sampler: Euler a OR DPM++ 2M/SDE Karras
Steps: 20-28
CFG: 11
Resolution: 512x768
High Res Fix: 1.5x, Latent (4x_foolhardy_Remacri), 0.5-0.7 Denosing
From there I img2img:
DDIM, 50 steps, resize OR latent upscale (4x_foolhardy_Remacri) to 960x1440, 0.3-0.7 Denosing (or less to keep more original details, usually more if using latent), CFG 7
Trained on Clip Skip 2, but 1 seems to work as well.
Throwing "bodysuit" in the prompt can sometimes help with the outfit.
Going to work on generating them for FE Engage characters since most models out there aren't trained on any of the new characters. Drop a comment if there are any you'd like to see next.