Recommendations (v2.0 - Anima)
Prompt: whitewash_eyes, white eyes, solid eyes, glowing eyes. Weight: 0.8 (0.6–1.0).
Other colors: replace white eyes, e.g. whitewash_eyes, pink eyes, solid eyes, glowing eyes. Keep solid eyes, glowing eyes, otherwise pupils can show through. Works best on portraits and upper body shots. Settings: 30 steps, CFG 4, er_sde / simple. More tips are in the version notes.
XL version of LORA, trained on model generation results Whitewash Eyes LoRA / Shirome / Hypnosis
Recomendations
Lora weight: There is no universal weight here, choose it to suit your needs. I'm using 0.5
You can set the eye color by replacing the white eyes token with another color. Although other colors may work less consistently
Description
Anima Base v1.0 version, trained locally (ai-toolkit) on 26 images.
Prompt: whitewash_eyes, white eyes, solid eyes, glowing eyes
Weight: 0.8 (0.6–1.0). It already works at 0.6 (see the weight comparisons).
Gives solid, evenly glowing eyes with no pupils or iris. Without the LoRA Anima draws a glowing iris with pupils for colored eyes, and only a small white circle in place of the iris for white eyes.
Tips:
Use the full combination. Without
solid eyescolored eyes can keep the pupils; withoutglowing eyeswhite eyes can keep the iris outline; with the trigger alone the model tends to draw closed eyes.Other colors: replace
white eyes, e.g.whitewash_eyes, pink eyes, solid eyes, glowing eyes.no pupilsalso works instead ofsolid eyes, but the color comes out paler; using both makes colored eyes almost white.Works on known characters (Ganyu, Raiden Shogun, Keqing, Yae Miko in the previews): the character design is kept, only the eyes change.
Portraits and upper body shots work best. On full-body shots the eyes are small: use a closer framing or a face detailer.
Pair with
expressionlessfor the blank, entranced look.
Tested with: 30 steps, CFG 4, er_sde / simple, 896×1152 and 1024×1024.
Negative: worst quality, low quality, score_1, score_2, score_3, blurry, jpeg artifacts, sepia, signature, watermark, text, artist name (add nsfw, nude for SFW).
DiT-only LoRA (no text encoder weights), rank 16 / alpha 16, 1500 steps.








