[Tested] 15 Ghost & Zombie Styling Prompts — Which Ones Turn a Beauty Inhuman Without Breaking Her?

TL;DR

Anything you can physically ADD to the look works. Zombie makeup (3/3), clouded eyes (3/3), vampire fangs and crimson eyes (3/3), doll joints (3/3), black tears (3/3), frost and visible breath (3/3). Makeup, props, liquids, frost — if it’s drawable, it lands.
Physically impossible states cannot be specified. A translucent ghost body (0/3) stayed fully opaque. Same limitation as “no reflection” (0/3) in our mirror test — absence can’t be prompted.
“Unnatural” expressions get normalized. An unnaturally wide smile (0/3) became just a pleasant smile. The “unnatural gets rounded off” rule from our pose test applies to facial expressions too.
For eroticism + horror in one frame: light zombie, torn dress, drowned look, or clouded eyes. Keeping the face beautiful and corrupting exactly one element is the most efficient formula.

Experiment Design

Model: z-image-turbo (8 steps / CFG 1.0 / 1024x1024)
Seeds: 42, 43, 44, fixed across all conditions
Base subject and stage: 1girl, 32yo japanese actress, white silk slip dress, long black hair, standing in a dim abandoned room at night, looking at viewer (only the yuki-onna moves to a snowy courtyard)
Variable: the inhuman-styling vocabulary only, per condition (15 conditions × 3 images = 45 images)

Our previous four horror tests (locations, mirrors, poses, fear escalation) were about place, composition, and context. This one is the final piece: turning the subject herself into something inhuman.

Zombie & Undead

Light zombie makeup (3/3)

Light zombie

1girl, 32yo japanese actress, subtle zombie makeup, pale skin, faint dark veins on her neck, white silk slip dress, long black hair, standing in a dim abandoned room at night, looking at viewer

seed 42	seed 43	seed 44

subtle zombie makeup rendered as dark-ringed eyes, pale skin, and faint neck veins in all three images. The face never breaks — it stops right at “sick beauty.” The easiest entry point into inhuman styling.

Heavy zombie (3/3 — the scariest of the set)

Heavy zombie

1girl, 32yo japanese actress, zombie, gray decayed skin, dark veins, cloudy eyes, torn dirty dress, disheveled long black hair, standing in a dim abandoned room at night, looking at viewer

seed 42	seed 43	seed 44

Gray skin, hollow eyes, blood at the lips, a shredded dress — everything landed in all three images, and even the room decayed further to match. Maximum horror of the 15 conditions. Yet the bone structure stays beautiful: this model consistently renders it as “a beautiful woman in zombie makeup” rather than actual decay.

Torn dress (3/3 — the ero slot)

Torn dress

1girl, 32yo japanese actress, torn tattered white slip dress, dirt smudges on her skin, long black hair, standing in a dim abandoned room at night, looking at viewer

seed 42	seed 43	seed 44

No supernatural element at all — this condition builds a “survivor who escaped something.” Tears and dirt landed 3/3 with zero exposure. An excellent compromise between horror context and allure.

Eye & Face Anomalies

Clouded pupil-less eyes (3/3 — best value per token)

Clouded eyes

1girl, 32yo japanese actress, pale cloudy white eyes without visible pupils, white silk slip dress, long black hair, standing in a dim abandoned room at night, looking at viewer

seed 42	seed 43	seed 44

Clouding succeeded in all three (seeds 42 and 44 near-total whiteout, seed 43 keeps a faint pupil). The face stays flawless while the eyes alone go inhuman. One phrase, maximum effect — the best efficiency in the set, and our cover image.

Black tears (3/3 — though they drift toward blood)

Black tears

1girl, 32yo japanese actress, black tears streaming from her eyes, white silk slip dress, long black hair, standing in a dim abandoned room at night, looking at viewer

seed 42	seed 43	seed 44

Tear streaks appeared 3/3, but the color spec black didn’t fully hold: 2 of 3 came out dark red-brown, closer to tears of blood. Seed 42 weeps from one eye only. A single streak down the cheek massively raises the uncanny factor — strong pick if you can tolerate the color drift.

Eyes hidden in shadow (0/3 — the shadow materializes as an object)

Shadowed eyes

1girl, 32yo japanese actress, her eyes hidden in deep black shadow, only her mouth visible, faint smile, white silk slip dress, long black hair, standing in a dim abandoned room at night, looking at viewer

seed 42	seed 43	seed 44

We wanted lighting — eyes sunk in deep shadow. Instead, all three images grew a physical black band over the eyes, like a sleep mask. The “shadow” we specified as a lighting state got rendered as an object. The faint smile did land 3/3, which makes it all the more frustrating. Combined with the failed hair-over-face attempts in our fear escalation test, hiding the face is consistently difficult with this model.

Unnaturally wide smile (0/3 — normalized away)

Unnatural smile

1girl, 32yo japanese actress, smiling unnaturally wide, eyes not smiling, white silk slip dress, long black hair, standing in a dim abandoned room at night, looking at viewer

seed 42	seed 43	seed 44

smiling unnaturally wide, eyes not smiling produced a perfectly pleasant smile three times out of three — eyes very much smiling. Our pose test showed “unnatural” pose adjectives get rounded into natural poses; the same law holds for expressions. What remains is the mismatch of a beaming smile in a ruined building — which is, admittedly, its own kind of creepy.

Japanese Yokai

Yuki-onna (3/3 — full-scene transformation)

Yuki-onna

1girl, 32yo japanese actress, yuki-onna, long white hair, frost on her eyelashes and skin, breath visible in cold air, white kimono, standing in a snow-covered courtyard at night, looking at viewer

seed 42	seed 43	seed 44

White hair, frosted lashes and brows, visible breath, snowy courtyard — every requested element landed in all three images. One word, yuki-onna, swaps subject, costume, and background in sync. The most complete transformation in the set.

Hannya mask (0/3 on the sculpt, 3/3 as “an eerie mask”)

Hannya mask

1girl, 32yo japanese actress, holding a hannya demon mask beside her face, white silk slip dress, long black hair, standing in a dim abandoned room at night, looking at viewer

seed 42	seed 43	seed 44

The composition — mask held beside the face — worked 3/3. The sculpt, however, is not a hannya: it came out as a fox-mask hybrid with horns and fangs (white face, red horns, golden eyes). A failure if you need an accurate hannya; perfectly serviceable as “hiding half her face behind something wrong.”

Kitsune mask (3/3 — more bewitching than scary)

Kitsune mask

1girl, 32yo japanese actress, white fox mask tilted on the side of her head, white silk slip dress, long black hair, standing in a dim abandoned room at night, looking at viewer

seed 42	seed 43	seed 44

A white fox mask appeared 3/3 (seed 42 leans cat-like), worn differently each seed: over one eye, on top of the head, hiding one eye. The mood is festival allure rather than fear, with the abandoned room adding just a whisper of wrongness. This is the “bewitching yokai” slot, not the horror slot.

Dolls, Vampires & States of Being

Ball-jointed doll (3/3 — the uncanny slot)

Ball-jointed doll

1girl, 32yo japanese actress, looking like a broken ball-jointed doll, visible doll joint lines on her arms, glass-like eyes, white silk slip dress, long black hair, standing in a dim abandoned room at night, looking at viewer

seed 42	seed 43	seed 44

Doll joints rendered on the arms in all three images. On seed 44 both arms become full doll mechanisms, hands included. Meanwhile the face stays entirely human (glass-like eyes barely registered). A living face on a doll’s body — this combination produced the single most uncanny image of the whole set.

Vampire (3/3 — the elegant inhuman)

Vampire

1girl, 32yo japanese actress, vampire, small fangs, crimson eyes, pale skin, elegant, white silk slip dress, long black hair, standing in a dim abandoned room at night, looking at viewer

seed 42	seed 43	seed 44

Crimson eyes, small fangs, blood traces at the mouth, pale skin — all landed 3/3. elegant held everything together with zero breakdown. The most refined way to go inhuman in this lineup.

Drowned look, soaked through (3/3 — maximum allure)

Drowned look

1girl, 32yo japanese actress, soaked waterlogged white dress clinging to her body, dripping wet long black hair, pale bluish skin, standing in a dim abandoned room at night, looking at viewer

NSFW - クリックで表示

seed 42	seed 43	seed 44

Clinging translucent fabric, dripping hair, bluish pale skin — all three landed. The sheerness is the highest in the set, and allure completely overpowers fear. If you want “something that climbed out of the well,” pair this styling with a water location from our location catalog and let the context supply the dread.

Translucent body (0/3 — the absence limit)

Translucent ghost

1girl, 32yo japanese actress, her body slightly translucent like a ghost, the room faintly visible through her, white silk slip dress, long black hair, standing in a dim abandoned room at night, looking at viewer

seed 42	seed 43	seed 44

The body stayed fully opaque in all three images; only a cold bluish grade came through. Just like “no reflection” (0/3) in the mirror test: absence and transparency simply cannot be specified with this model.

Visible breath (3/3 — but it becomes “smoke”)

Visible breath

1girl, 32yo japanese actress, pale skin, breath visible as white mist in the cold air, goosebumps on her arms, white silk slip dress, long black hair, standing in a dim abandoned room at night, looking at viewer

seed 42	seed 43	seed 44

We asked for breath; we got smoke-like haze drifting beside and behind her head in all three images (goosebumps: 0/3). Contrast with the yuki-onna (z06), where breath appeared naturally in a snow scene — without a cold context, “breath” mutates into “smoke.” Read charitably as “mystery haze in an abandoned room,” it’s a decent accident.

Note: Clouded eyes produced a major shift in impression from a single phrase. The face itself remains the same as always, but eye contact reads as distinctly not-human. The most unsettling result of this test was doll seed 44 — a living face paired with doll-like arms.

Summary: The Three Laws of Going Inhuman

Three laws emerged from 15 conditions and 45 images:

Additive looks work. Makeup (zombie, dark circles), props (masks, fangs), liquids (tears, blood, water), frost and snow — physically drawable additions passed 3/3.
Physically impossible states can’t be specified. Translucent body (0/3), no reflection (0/3, mirror test). Absence and transparency are currently off the table.
“Unnatural” descriptors get normalized. The unnatural smile (0/3) became a pleasant one. Same behavior as the pose test: build fear out of concrete objects, not adjectives.

Recommendations by goal:

Goal	Styling
Eroticism + horror	Light zombie / torn dress / drowned look / clouded eyes
Maximum fear	Heavy zombie
Full transformation	Yuki-onna
Uncanny wrongness	Ball-jointed doll
Refined inhuman	Vampire
Bewitching, Japanese-style	Kitsune mask

Note: This completes the horror×ero pentalogy toolkit — casting (inhuman styling), stage (locations), props (mirrors), acting (poses), and direction (fear vocabulary). A combined finale covering all five elements is planned next.

The Setup Behind This Test

No local GPU? A cloud GPU lets you run bulk experiments like this one (45 images) in no time.

PR RunPod クラウドGPUでAI画像生成 RunPodを始める →