Conclusions First
- To keep the subject glamorous while making the scene scary, change the location. Swapping the room for an abandoned hospital left the subject’s face and the dress’s sheen at control level while only the background decayed (3 of 3 images).
- Adding
horror movie still+eerie unsettling atmosphereis the most powerful whole-image transform. Hair turned into wet strands, the color grade went cold and desaturated, and the expression went blank (each in 3 of 3 images) — without touching a single hair or expression keyword. The trade-off: the subject’s glamour goes down with it. white burial kimonoalone does not produce a burial kimono. All 3 images rendered an elegant white kimono with a decorative obi.slipping off one shoulderreproduced in 0 of 3 images.lit from below by flashlightdoes not reproduce the classic campfire-ghost-story underlighting. Instead, an actual flashlight tends to materialize in frame (1 of 3 in this condition; in the combined condition the subject held a lit flashlight in 2 of 3 images).
Motivation
The woman in white from Japanese horror films is terrifying, yet you can’t look away. Can z-image-turbo recreate that mix of dread and allure? We decompose “scariness” into four axes — location, lighting, costume (styling), and style vocabulary — and test one axis at a time to see which axis adds fear and which one destroys the allure.
Experiment Setup
| Item | Value |
|---|---|
| Model | z-image-turbo (6B, photorealistic distilled) |
| Steps | 8 |
| Sampler | euler / ddim_uniform |
| CFG | 1.0 |
| Size | 1024x1024 |
| Seeds | 42, 43, 44 (fixed) |
| Per condition | 3 images |
| Total | 6 conditions × 3 = 18 images |
Base Prompt (Control)
A white silk slip dress in a moonlit Japanese room — we add one “scary” axis at a time on top of this base.
Conditions
| Condition | Axis | Change |
|---|---|---|
| h00: Control | — | Base above |
| h01: Location | Place | Japanese room → abandoned hospital corridor + broken windows |
| h02: Lighting | Light | Moonlight → lit from below by flashlight, deep shadows on face |
| h03: Ghost styling | Costume & hair | Slip dress → white burial kimono slipping off one shoulder + long disheveled black hair partly covering her face |
| h04: Horror vocabulary | Style words | horror movie still prepended, eerie unsettling atmosphere appended |
| h05: Everything | Combined | All of the above at once |
Note: h03 changes costume and hair together as a styling set, and h05 is a combined condition. For these two we only observe the effect of the element group and make no claims about individual elements.
Results by Condition
h00: Control
| seed 42 | seed 43 | seed 44 |
|---|---|---|
 |  |  |
Observations: All 3 images show a woman in a glossy slip dress standing in a moonlit Japanese room with shoji screens. The face is polished with clearly rendered lip makeup. Nothing reads as scary. This is our reference point: allure 100, fear 0.
h01: Location (Abandoned Hospital)
| seed 42 | seed 43 | seed 44 |
|---|---|---|
 |  |  |
Observations: 3 of 3 images reproduced a convincingly derelict hospital corridor — shattered window panes, peeling pale-green wainscoting, debris scattered on the floor. Meanwhile the subject is virtually identical to control: no degradation in facial rendering, dress sheen, or figure. Seed 43 shortened the dress to mini length (the other two are midi). Seed 44 glitched, drawing the moon near the corridor ceiling.
Only the background gets scary; the subject is untouched. This was the most reliable way to get “a beautiful woman standing in a terrifying place.”
h02: Lighting (Flashlight from Below)
| seed 42 | seed 43 | seed 44 |
|---|---|---|
 |  |  |
Observations: The classic ghost-story underlighting — face lit from below with inverted shadows — failed to appear clearly in 3 of 3 images. Seed 42 got a pool of light on the neck and chin but a normally lit face. Seed 43 turned the face pale bluish and materialized an actual flashlight beaming in the lower right corner. Seed 44 is dim overall with a pale face but no shadow inversion.
Even with deep shadows on face, the model showed a tendency to keep the subject’s face visible. As a side effect, a pale, unhealthy complexion appeared in 2 of 3 images (seeds 43, 44).
h03: Ghost Styling (Burial Kimono + Disheveled Hair)
| seed 42 | seed 43 | seed 44 |
|---|---|---|
 |  |  |
Observations: A big miss versus expectations. Despite specifying white burial kimono, 3 of 3 images produced an elegant white kimono with a decorative floral obi. None of the burial-garment features (plain fabric, no obi, right-over-left wrap) appeared. slipping off one shoulder reproduced in 0 of 3 — no shoulder exposure at all.
Meanwhile, long disheveled black hair partly covering her face worked in 3 of 3, with damp-looking strands flowing over the face and shoulders. The result is not a scary ghost but a kimono beauty with post-bath hair: fear near zero, allure high but in a different direction from control.
h04: Horror Vocabulary (horror movie still)
| seed 42 | seed 43 | seed 44 |
|---|---|---|
 |  |  |
Observations: The largest change in the whole experiment. We added only two style phrases, yet all of the following changed in 3 of 3 images:
- Color grade: cold, desaturated, low-contrast blue-gray
- Hair texture: turned into wet-looking straight strands clinging to the face and body — with zero hair keywords changed
- Expression: went blank, with darker shadows around the eyes
- Figure: rendered thinner than control, with sharper collarbones and shoulders
- Composition: shifted to frontal, static, symmetrical framing
Additionally, seed 43 added stains to the dress and seed 44 gave it a dingy cast (seed 42’s dress stayed clean). The “horror movie still” context consistently rewrites hair, skin, wardrobe, and pose.
h05: Everything Combined
| seed 42 | seed 43 | seed 44 |
|---|---|---|
 |  |  |
Observations: A full J-horror frame. The kimono, decorative in h03, went plain and burial-garment-adjacent in 3 of 3 images here (being a combined condition, we can’t attribute this to any single element). Hair dishevelment is the highest of all conditions, dark circles appear under the eyes, and the face is pale. In seeds 42 and 44, the subject herself holds a lit flashlight in the corridor — lit from below by flashlight got interpreted as “a person holding a flashlight.”
Fear factor is unmatched, but skin exposure is zero and the body-line rendering of the glossy dress from control is gone. This condition makes the fear-versus-allure trade-off explicit.
Cross-Condition Comparison (seed 42)
Lining up all 6 conditions at the same seed (42) shows the gradient clearly.
| Control | Abandoned hospital | Underlighting |
|---|---|---|
| | |
| Ghost styling | Horror vocabulary | Everything |
|---|---|---|
| | |
Location and vocabulary contribute most to “scariness,” while standalone lighting and costume keywords resist working in the intended direction — a clear asymmetry.
Summary: A Recipe for Horror-Tinged Allure
What the 18 images taught us:
| Desired image | Recommended approach |
|---|---|
| Beautiful subject in a scary place | Change only the location (abandoned hospital corridor, etc.). Subject rendering survives intact |
| Turn the whole frame into a horror film | horror movie still + eerie unsettling atmosphere. Hair, expression, color grade, and wardrobe all convert at once |
| Keep the allure | Skip the horror vocabulary — it dulled the subject’s glamour in 3 of 3 images |
| Get a burial kimono | white burial kimono alone is not enough (0 of 3). It only went plain when combined with horror vocabulary (observed in the combined condition) |
| Ghost-story underlighting | lit from below by flashlight does not reproduce it. Watch out for the side effect of an actual flashlight appearing |
Note these results are specific to z-image-turbo (CFG=1.0, 8 steps); other models and parameters may behave differently. For prompt design fundamentals, see our verified best practices.
![[Verified] Image Generation Prompt Best Practices](/tips/prompt-best-practices/cover_0_0000_4517457392071889496.webp)
![[Experiment] Prompts for Goth and Dark Makeup | 8 Conditions, 24 Images](/tips/goth-dark-makeup-test/a01_goth_makeup_s2_0_0000_7295072554507705269.webp)
![[Experiment] Night Light Source Comparison — How Do Neon, Street Lamps, and Moonlight Change Skin Color?](/tips/night-light-source-comparison/cover.webp)