A character LoRA for Valeriy, Major of the Fatui Ordnance Palace and leader of the 1st Guards Grenadier Regiment's special task battalion from Genshin Impact — trained to nail his short black hair, purple eyes, hair between eyes and ahoge, plus the fur-trimmed black coat, shoulder armor and black gloves. Both his Snezhnayan military uniform and a modern suit look are covered.

rigger word valeriy (genshin impact) — put it immediately after your quality tags. Anima's built-in tag list does not know this character yet, so the trigger is mandatory; without it you get a generic dark-haired man.

Recommended settings Strength 0.8–1.0 (1.0 for full costume fidelity, 0.8 when blending with another character or a style LoRA) · Sampler er_sde, scheduler beta57 or simple · Steps 30–40 · CFG 4–5 · Resolution 832×1216 portrait / 1024×1024 square (1152×1536 + hires for detail) · Base model Anima base v1.0.

What it's good at

• Character fidelity — every one of the 156 training captions carried his identity tags: short hair black hair purple eyes hair between eyes ahoge.

• Costume — fur trim ×120, gloves ×52, coat ×46, shoulder armor ×31, long sleeves, black gloves, cape. The Snezhnayan uniform is his strongest suit; black suit + necktie + white shirt civilian looks also work.

• Portraits & expressions — upper body ×76, looking at viewer ×94, closed mouth ×91, parted lips ×19, blush ×15, smile ×15, expressionless ×9, closed eyes ×11.

• Group & scene — multiple boys ×44, 2boys ×43, yaoi ×16, 1girl ×17; outdoors with snowing ×9, sky, blue sky, cloud, day; muscular ×9 / pectorals ×8 for a beefier build.

Usage notes

• This is a character, not a style LoRA. If you stack a style LoRA on top, keep the style at 0.5–0.6 or it will overwrite his identity.

• Long prompts dilute the trigger: past ~50 tags the trigger's influence drops — raise this LoRA's strength or move the trigger earlier.

• Keep his identity tags in the prompt; drop them and the face drifts. ahoge appears in nearly every training caption — keep it if you want the hair silhouette.

Training details

• 156 images × 2 repeats × 14 epochs = 4,368 image-passes, 1,092 steps (batch 2 × grad-accum 2)

• Network dim 32 / alpha 16, UNet only, fp16 · Optimizer Prodigy (lr 1, constant, decouple + bias correction)

• 1024 buckets (512–1536), anima_baseV10, keep_tokens 1, caption shuffle + 5% tag dropout · Danbooru-style tags, trigger pinned first