Posting two models because after extensive testing, I think they both have pros and cons. See some of the images for comparisons.<ul><li>v1.0 seems to adhere more to the Arcane style at full weight if the prompts are kept simple and close to the training data. Also seems to be slightly better with NSFW.</li><li>v1.0-offset seems to be much more versatile at a lower weight (0.65 to 0.8). I'd recommend using this one for just about anything that's not recreating stills from Arcane</li></ul>Some keywords to use in prompts are: "arcane, vi, red hair, light blue eyes, red jacket, nose piercing, face tattoo, vi (league of legends), eyebrow scar, ear piercing, bandaged arms". Use "portrait", "profile", "upper body", "full body", or "from behind" to help with composition and orientation.

Arcane Vi SDXL

Offset version created alongside initial version. Much more versatile at lower weights. Same prompting rules as v1.0 apply.

Files with SHA256: 2d92af994a65f9df746bde88ab5d2b33fb3bd36e0e766f46628bd40f110f0554