Key Findings
- Shiro gyaru (light-skinned gyaru) was reproduced in 3/3 images with the recipe
gyaru, platinum blonde hair, fair skin, glossy pink lips, false lashes. Platinum blonde long hair, fair skin, glossy lips, and long lashes appeared consistently - Kuro gyaru (tanned gyaru) was reproduced in 3/3 images with
gyaru, deep tanned skin, bleached blonde hair, white eyeshadow. The tan + bleached blonde + white eyeshadow combination rendered clearly - Bare
gyarudiffers little from control, matching what we found in the gyaru makeup test, now confirmed in an upper-body composition - The Y2K recipe
y2k gyaru, 2000s japanese street fashion, tanned skinproduced a gyaru look in 0/3 images. Y2K-style casual clothing (distressed denim shorts, logo tank tops) appeared, but hair, makeup, and skin stayed close to control - Applying the shiro/kuro recipes also changes facial identity substantially as a side effect (which keyword within the recipe causes this was not isolated)
Experiment Design
| Item | Value |
|---|---|
| Model | z-image-turbo (6B, photorealistic distilled) |
| Steps | 8 |
| Sampler | euler |
| Scheduler | ddim_uniform |
| CFG | 1.0 |
| Size | 1024x1024 |
| Seeds | 42 / 43 / 44 fixed (shared across conditions) |
| Per condition | 3 images (3 seeds) |
| Total | 5 conditions = 15 images |
Base Prompt
We used an upper-body street portrait rather than a face close-up so that makeup, hair, skin, and clothing can all be observed at once.
Caveat: The shiro, kuro, and Y2K conditions are multi-keyword “recipes,” not single-variable changes. Conclusions therefore apply to each recipe as a whole; individual keyword contributions cannot be isolated.
A00: control (no style keywords)
| seed 42 | seed 43 | seed 44 |
|---|---|---|
![]() | ![]() | ![]() |
Observations: All three images show dark straight hair with natural makeup. Outfits are subdued office-casual (gray tops, tweed jackets). No gyaru elements.
A01: gyaru (single keyword)
| seed 42 | seed 43 | seed 44 |
|---|---|---|
![]() | ![]() | ![]() |
Observations: Hair stays dark and makeup stays natural in all three images — little difference from control. Bangs appear and the subject looks somewhat younger, but gyaru-style changes to hair color, skin, or makeup occurred in 0/3 images. Consistent with the face close-up results in the gyaru makeup test: bare gyaru is weak.
A02: Shiro gyaru recipe
| seed 42 | seed 43 | seed 44 |
|---|---|---|
![]() | ![]() | ![]() |
Observations: 3/3 images show platinum blonde long hair, fair skin, glossy pink-to-red lips, and long dense lashes. Eyeliner is clearly defined, producing a stable shiro-gyaru look. Outfits also shift toward skin-showing styles (camisoles, V-neck knits). However, the face itself changes substantially versus control — it no longer reads as the same person.
A03: Kuro gyaru recipe
| seed 42 | seed 43 | seed 44 |
|---|---|---|
![]() | ![]() | ![]() |
Observations: 3/3 images render tanned skin, bleached blonde long hair, and white eyeshadow together. In the gyaru makeup test, bare ganguro did not produce tanned skin; this recipe with an explicit deep tanned skin does. Outfits also shift to revealing street styles (crop tops, low-rise denim).
A04: Heisei gyaru (Y2K) recipe
| seed 42 | seed 43 | seed 44 |
|---|---|---|
![]() | ![]() | ![]() |
Observations: 0/3 images show gyaru-style hair or makeup. Hair stays dark, and despite tanned skin, clearly tanned skin appeared in 0/3 images. Clothing does change to Y2K-flavored casual — distressed denim shorts, logo tank tops, wide belts — so “2000s outfit” is reflected, but the look falls short of Heisei gyaru. Logo shirt text renders as unreadable artifacts.
Note: Based on these results, a promising route to Heisei gyaru would be adding Y2K clothing keywords to the kuro gyaru recipe (A03). Concrete hair/skin/makeup keywords, as in A02/A03, work more reliably than an abstract word like
y2k.
Summary
| Condition | Hair change | Skin change | Makeup change | Outfit change | Success |
|---|---|---|---|---|---|
| control | - | - | - | - | - |
| gyaru only | none | none | none | slight | 0/3 |
| Shiro recipe | platinum blonde | fair | lips/lashes emphasized | skin-showing | 3/3 |
| Kuro recipe | bleached blonde | tanned | white eyeshadow | street style | 3/3 |
| Y2K recipe | none | none | none | Y2K casual | 0/3 |
Rather than abstract style names like gyaru, stacking concrete keywords for hair color, skin, and makeup parts proved effective for controlling gyaru sub-styles. This matches the “stacking parts is more reliable” conclusion from the gyaru makeup test.
Note: In both the shiro and kuro recipes, applying the recipe also changed the face to a different-person level. These recipes are not suitable when you want to keep the same character and only change the style.















![[Experiment] Prompts for Gyaru Makeup | 8 Conditions, 24 Images](/tips/gyaru-makeup-test/a05_false_lashes_s2_0_0000_7295072554507705269.webp)

![[Verified] Image Generation Prompt Best Practices](/tips/prompt-best-practices/cover_0_0000_4517457392071889496.webp)