This LoRA was trained on a broad selection of images from the game, aiming to bring its distinctive style to newly generated pictures.
The dataset was composed of around 270 images at 896x1152 resolution.
Illustrious
It tends to work best when given some creative freedom. So far, 896 × 1152 appears to be the optimal resolution, and a LoRA weight from 0.8 to 1.2 seems to offer the best balance.
All showcase images were generated at 896 × 1152, then upscaled by 2× with Hi-Res Fix and enhanced with Face Detailer.
Anima
It's the first time I train an Anima Lora and the first Lora trained on my machine!
896×1152 appears to be again the optimal resolution, weight should be around 1.2.
Showcase images were generated with the above resolution at 14 step using a Turbo Lora and upscalig at 1.5x using Hi-Res Fix with 2 or 3 more steps.