Character trigger word: Denia

Character and style rendering are inseparable here. What I like is this character in this particular style, not the same character rendered in some other style. The training data only uses in-game photography of the character with adjustable graphics settings, aiming to reproduce the game’s stylized rendering as closely as possible.

This version of the LoRA has some issues with pose accuracy. I’ll need to run a few more rounds of testing with natural-language descriptions to see whether it can be corrected. If using VNCCS can restore the pose properly, then there’s no need to worry about the LoRA itself.

After all, this is only the first step in the ComfyUI workflow. Other shortcomings can be addressed through other methods later.