Welcome to my first LoRA!
Basic Parameters:
CFG Scale: 7
Steps: 20
Sampler: DPM++ 2m
Base Model: WAI-illustrious-SDXL (Illustrious)
Hires upscaler: R-ESRGAN 4x + Anime6B
Adetailer for face and hands
Trigger Words:
The main trigger word is "Tomioka Giyuu", and in my case I always add style and quality terms to the prompt such as:
"anime screencap, anime source, anime coloring, masterpiece, best quality, high resolution, 4K, 8K, colorful."If you want water streams to appear around Tomioka in the image, add:
"battle stance, surrounded by streams of water" to the prompt.If you want him to appear holding an unsheathed sword, add:
"holding unsheathed katana".If instead you want him to appear with the sword sheathed on the belt along with the scabbard, add:
"unsheathed katana, katana sheath on the belt".
Things to Keep in Mind:
The LoRA is fully trained on the character’s official anime outfit, so you don’t need to specify anything about the clothing if you want him to appear in his main outfit—unless you want him to wear something different.
To be honest, the sword often shows visual glitches, as do the hands when holding it, or the belt area where the scabbard should be—especially in distant full-body shots. If you try multiple generations, you’ll eventually get a good image with the sword. But if you don’t want the sword at all, I strongly recommend specifying it in the negative prompt. Likewise, it often generates more than one sword, so it’s helpful to include negative terms like:
"more than 1 katana, 2 katanas", etc.
(Sorry, it’s my first LoRA and I haven’t figured out how to fix these details yet.)
Have fun!