LORA based on front-facing portraits of all released characters of Genshin Impact (as of version 3.4).

General prompt recipe is simple background, portrait, solo, plus either 1girl or 1boy. There are no trigger words.

LORA was trained against Anything v3. All examples were generated with Anything v3 with a LORA weight of 1.0 at 512x512 plus 1.5x upscaling using hi-res fix.