Rias Gremory from High School DxD, with a realism-focused approach.
I made this LoRA with one main goal in mind: take Rias and push her toward a more realistic rendering while keeping the recognizable character features intact.
And honestly, I think the results speak for themselves.
The LoRA is particularly focused on realistic hair rendering, skin, eyes, fabric/material detail, lighting, and overall photographic depth. Rather than simply making the character look "semi-realistic," I wanted the outputs to actually have that feeling of a properly rendered person — individual hair strands, subtle skin texture, believable highlights, detailed clothing materials, etc.
I've tested it across quite a few different outfits, poses, compositions and lighting conditions, and the results have been surprisingly consistent.
The examples here were generated on the same base model recommended below, so if you're wondering "okay, but can I actually reproduce this?" — yes, that's the idea.
Recommended Settings
Base Model: HassakuXlIllustrious V22
LoRA Weight: 0.8
Steps: 35
CFG: 5–7
I'd recommend sticking reasonably close to these settings, especially the base model. Since this LoRA was trained and tested specifically around HassakuXlIllustrious V22, that's where I found the most consistent results.
You can obviously experiment with other checkpoints and weights — that's half the fun — but the images shown here are generated on the same model, so they represent what you can expect from the recommended setup.
What I was aiming for
Realistic skin rendering
Individual hair strands and detailed hair highlights
More natural-looking eyes
Detailed fabric and clothing materials
Stronger lighting and material response
Realistic depth and shading
Retaining Rias' recognizable facial features and overall appearance
Flexibility across different outfits and compositions
Basically: Rias, but rendered like an actual person rather than just slapping a realism filter over an anime character.
Give it a try, play around with the weight, and see what you can get out of it. The examples above are probably the best way I can demonstrate what the LoRA is capable of.