[Experiment] Prompts for Light/Pink Areola | Thorough Testing with 60 Images

Conclusions First

Controlling light areola color is generally more difficult than controlling dark areola color. Testing 20 conditions and 60 images, no condition achieved an areola “nearly the same color as the skin.”

In the dark areola test, chocolate areola achieved high contrast in just 2 tokens, but no equivalent “decisive” prompt was found for the lighter direction.

Prompts with Weak but Confirmed Effect

Prompts Not to Use

• strawberry areola — Areola turns vivid red and a fruit object is drawn on top of the nipple
• cherry blossom areola — Cherry blossom petal shapes are drawn on the areola
• white areola — White nipple-cover-like artifacts appear

Experiment Design

Base Prompt

Same base prompt as the dark areola test.

A. Control (no variable)

NSFW - クリックで表示

seed 1	seed 2	seed 3

Skin tone is ivory, areola is light brown. Contrast is medium. This is the baseline.

B. Color Adjective × areola

b01: light areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

Same light brown as control. No change.

b02: pale areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

Same as control. No change.

b03: pink areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

2 out of 3 show a slight pink tint added to the areola. However, brightness change (lighter direction) is almost absent — just a slight shift in hue toward red. Pink/red color bleeding into the background in seed 3.

b04: light pink areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

Same degree as pink areola. Adding “light” shows no observed lightening.

b05: white areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

No areola color change. Instead, 2 out of 3 show a white circular object (nipple-cover-like) drawn on the nipple as a side effect. Should be avoided.

Group B Summary

Color adjectives alone cannot lighten areola color.

C. Color Adjective × nipples

c01: pink nipples

NSFW - クリックで表示

seed 1	seed 2	seed 3

Equivalent to pink areola. No clear pink, just slight hue shift.

c02: pale nipples

NSFW - クリックで表示

seed 1	seed 2	seed 3

2 out of 3 show the areola shifting slightly toward skin tone with somewhat reduced contrast. More effect than Group B but not dramatic. Text artifacts appear extensively in seed 2.

D. Skin Tone + Areola Color Combinations

d01: fair skin, pink areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

Same as B03 (pink areola). Adding fair skin keeps skin tone white but has no additional effect on areola color.

d02: fair skin, light pink areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

2 out of 3 show a slight tendency for the areola to move toward skin color. However the difference is minimal.

d03: pale skin, pink areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

Not much different from d01/d02. pale skin shifts skin slightly toward a bluish-white direction.

Group D Summary

Skin tone specification combinations barely contribute to areola color lightening.

E. Concrete Object Metaphors

In the dark areola test, chocolate areola was most effective. Trying the same approach with light-colored objects.

e01: fair skin, peach areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

All 3 stable at light brown with low-to-medium contrast. Tendency to be slightly lighter than Group B and D. Side effects are relatively small. Best balance in Group E.

e02: fair skin, rose areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

No lightening effect. Rather, a slightly darker brown with redness added. Prominent redness spreading to the cheek in seed 1. Floral patterned background illustration appears in seed 3 — the “rose” semantics bleeding into the composition.

e03: fair skin, strawberry areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

Completely counter-productive. All 3 change the areola to vivid red with very high contrast. Seed 1 has a severe side effect where an actual strawberry is drawn on top of the nipple. Forbidden.

Lab Director Comment: Why though? A strawberry literally sitting on the nipple is terrifying. chocolate works as just color but strawberry comes through as a physical object — that’s the trap of concrete object metaphors.

e04: fair skin, salmon areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

Same degree as peach. Light brown to pinkish with low-to-medium contrast. Nearly the same result as peach.

e05: fair skin, cherry blossom areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

Weak effect toward light pink direction, but 2 out of 3 show a serious side effect where cherry blossom petal shapes are physically drawn on the areola. The concrete object metaphor is interpreted as an “object” rather than “color.” Should be avoided.

Group E Summary

The success of chocolate was not reproduced in the lighter direction. Concrete objects with a clearly defined shape (strawberry, cherry blossom) risk being rendered as objects rather than colors. peach/salmon have an ambiguous shape (also recognized as color names), so only the color information gets through without objectification.

F. Abstract Descriptions

f01: fair skin, barely visible areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

Same as control. “barely visible” is not interpreted as an areola color attribute.

f02: areola same color as skin

NSFW - クリックで表示

seed 1	seed 2	seed 3

No effect, contrary to expectations. Seed 2 actually has darker brown with increased contrast. Natural language semantic instructions (“make it the same color”) don’t get through to the model.

f03: fair skin, faded areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

Same as control. The meaning of “faded” is not being communicated to the model.

f04: fair skin, translucent areola

NSFW - クリックで表示

seed 1	seed 2	seed 3

Best in Group F. All 3 show low contrast with areola close to skin-tone light brown/light pink. Weak effect comparable to peach/salmon.

Group F Summary

Abstract semantic instructions don’t get through to the model. Only translucent shows a weak effect, comparable to the concrete metaphors peach/salmon.

Full Condition Summary

Lab Director Comment

So, dark areola was easy with chocolate, but 60 images of testing the light direction and finding no decisive prompt — the results speak for themselves. peach areola is the best we’ve got but the effect is weak. The asymmetry of “adding and darkening = easy” vs “removing and lightening = hard” seems like a characteristic of the model worth remembering.

Practical Recommendations

Safest Prompt

• Most stable lightening in 2 tokens
• No objectification risk
• However, the effect is weak and won’t achieve “nearly the same color as skin”

Asymmetry Between Dark and Light Areola

The darker direction had a stronger effect. The reason is unknown, but this asymmetry was consistently observed.

What Not to Do

• light / pale / faded don’t work — abstract light-color adjectives don’t get through to the model
• Don’t use white areola — nipple-cover-like artifacts appear
• Never use strawberry — dual side effects of vivid reddening + fruit drawing
• Also avoid cherry blossom — petals are drawn on the areola
• Natural language semantic instructions (same color as skin) don’t work

Why the Difference in Concrete Object Metaphors?

Thinking with the same principle that made chocolate work in the dark areola test:

• chocolate — Ambiguous shape (liquid/solid/powder) → only color information transmitted → success
• peach / salmon — Ambiguous shape (also recognized as color names) → only color information transmitted → weak success
• strawberry — Clear shape (red fruit with seeds) → rendered as object → failure
• cherry blossom — Clear shape (5 petals) → rendered as object → failure

When using concrete object metaphors, choose words with an ambiguous form that are also used as color names.

Related Articles

[Verified] Image Generation Prompt Best Practices

[Experiment] Prompts for 'Dark Areola on Fair Skin' | Thorough Verification with 54 Images