Hush from VALORANT, a character LoRA for Illustrious XL / WAI-NSFW Illustrious (v17).
Trigger: hush valorant
Trained to reproduce her default outfit faithfully: very long hair fading from white-lilac to violet with blunt hime bangs, magenta eyes with a teal diamond highlight, a small black diamond mole under her left eye, black cropped high-collar jacket with an ornate round brooch, black wrap crop top, asymmetric wrap skirt with a geometric pattern over a burgundy layer, black tights, white tactical gloves and heeled ankle boots. Add "floating pink shards, floating dice" for her telekinesis effects.
Recommended weight: 0.8 - 1.0. Standard Illustrious/WAI settings work fine (the showcase images use Euler a, 28 steps, CFG 7).
Made by Offcanon. I accept character and style LoRA commissions (fan works for your personal use; IP rights are the client's responsibility) — DM me here if you're interested.
Description
[v5 — full redesign] New trigger word: hxsh
IMPORTANT: v5 uses the single trigger word "hxsh" (the old "hush valorant" trigger does NOT work for v5). Do NOT add the word "valorant" anywhere in your prompt with v5 — it pulls the generation toward a different character and breaks the face/hair.
What changed vs v3: the face was rebuilt from the official art as ground truth (new coined trigger, curated face-anchor dataset, LoCon training). The face now matches the official design much more closely: sharp heavy-lidded eyes, compact proportions, the single diamond mole under her left eye, magenta iris with the teal segment, fair skin.
Recommended weight: 0.8–1.0 (usable range 0.6–1.2; identity already holds at 0.6). Trained on WAI v17 (Illustrious). The trigger alone summons the canonical outfit; outfit changes, hairstyle changes (e.g. short bob), and style prompts (manga / anime screencap / flat / watercolor) all work.
Verification notes (fixed-seed test set): anatomy breakdowns 0/18 at weight 1.0 including seated and long-dress poses (v4.1 had 4/18). Known remaining issues: the mole occasionally doubles (about 1 in 8 close-ups at 0.8); chibi prompts only partially deform; qipao-type prompts can darken the hair color; a held prop may rarely duplicate.