[Experiment] Prompts for a No-Makeup / Natural Look | 8 Conditions, 24 Images

[Experiment] Prompts for a No-Makeup / Natural Look | 8 Conditions, 24 Images

Conclusion First

  • z-image-turbo’s default output already has light makeup (faint lip color, eyeliner, even skin tone)
  • bare face was the most effective at reducing makeup. Lips shift toward beige tones, skin becomes less uniform and closer to natural
  • no makeup shows almost no difference from the control — weak makeup removal effect
  • natural makeup, minimal makeup, fresh face, nude makeup, and skin care all show minimal difference from the control — insufficient for a true no-makeup look
  • If aiming for a completely bare face, bare face is recommended, but even then fully removing the default makeup is difficult

Experimental Design

ItemValue
Modelz-image-turbo (6B, photorealistic distilled)
Steps8
Samplereuler
Schedulerddim_uniform
CFG1.0
Size1024x1024
Seeds3 fixed seeds (common across all conditions)
Per condition3 images (3 seeds)
Total8 conditions = 24 images

Base Prompt

Base Prompt
1girl, 32yo japanese actress, {VARIABLE}, close-up face portrait, indoor studio, soft lighting

Using close-up face portrait to observe makeup presence/absence.

A00: control

seed 1seed 2seed 3
control s1control s2control s3

Observation: The default output already shows light makeup. Lips have a pink-to-coral color, eyes show faint eyeliner, skin is uniformly smooth. Seed 3 features a necklace and veil-like fabric, giving a somewhat more glamorous impression. Overall, the baseline is “an actress with natural makeup.”

A01: no makeup

Added Keyword
no makeup
seed 1seed 2seed 3
no makeup s1no makeup s2no makeup s3

Observation: Almost no difference from the control. Lip color, skin uniformity, and eyeliner all remain at similar levels to the control. Comparing seed 2, the composition, expression, and makeup level are nearly identical — no makeup is not functioning as a makeup removal instruction.

A02: bare face

Added Keyword
bare face
seed 1seed 2seed 3
bare face s1bare face s2bare face s3

Observation: The condition with the largest change in this experiment. Seed 2 shows lips shifting to beige tones — clearly different from the control’s pink-to-coral. Skin is more matte with the gloss of the control suppressed. Eyebrow outline is also slightly softer, and overall makeup feel is reduced. However, the change is more like “just after removing light makeup” rather than a completely bare face.

Lab Director Comment: “Just after removing light makeup” — honestly that’s not really no-makeup is it. The default makeup feel is just too strong.

A03: natural makeup

Added Keyword
natural makeup
seed 1seed 2seed 3
natural makeup s1natural makeup s2natural makeup s3

Observation: No difference from control. Pink lip color, uniform smooth skin, faint eyeliner — all at control levels. Since the default output is already effectively “natural makeup” equivalent, adding this keyword changes nothing.

A04: minimal makeup

Added Keyword
minimal makeup
seed 1seed 2seed 3
minimal makeup s1minimal makeup s2minimal makeup s3

Observation: Barely any difference from control. Comparing seed 2, lips may be very slightly paler than control, but this cannot be claimed as a reproducible difference. Skin texture and eye makeup are at the same level as control. Like natural makeup, the difference from the default output is too small to be practically useful.

A05: fresh face

Added Keyword
fresh face
seed 1seed 2seed 3
fresh face s1fresh face s2fresh face s3

Observation: No difference from control. Lip color, skin uniformity, and eye makeup are all at control levels. fresh face appears to be interpreted as “clean-looking face” rather than absence of makeup.

A06: nude makeup

Added Keyword
nude makeup
seed 1seed 2seed 3
nude makeup s1nude makeup s2nude makeup s3

Observation: Seed 2 shows a beige scarf/cloth appearing on the head — the word nude may have influenced clothing or accessories. The makeup itself shows little difference from the control. As a makeup genre, “nude makeup” means “skin-tone makeup,” but in the model’s output it’s essentially the same as control, with the risk of nude’s semantic interference causing unintended side effects on accessories.

A07: skin care

Added Keyword
skin care
seed 1seed 2seed 3
skin care s1skin care s2skin care s3

Observation: No difference from control. Lip color, skin texture, and eye makeup are all at control levels. skin care is a keyword related to skin condition rather than presence or absence of makeup — no makeup removal effect.

Summary

ConditionMakeup Removal EffectNotes
bare facePresentLips shift to beige, skin becomes more matte
no makeupNoneNo difference from control
natural makeupNoneDefault output is already equivalent
minimal makeupAlmost nonePossible slight difference but reproducibility uncertain
fresh faceNoneActs on facial impression, not makeup
nude makeupNoneRisk of nude semantic interference changing accessories
skin careNoneNo effect on makeup

z-image-turbo’s default output includes light makeup, and removing it is not easy. bare face is the most effective, but even then it doesn’t reach a truly bare face. If you want to reduce makeup further, combining bare face with expressions denying individual makeup elements like no lipstick, no eyeliner may be worth exploring — though note that negation phrases (no X) in positive prompts tend not to be accurately interpreted as negation by the text encoder.

Lab Director Comment: So, I didn’t expect getting a no-makeup look to be this hard. It’s interesting that adding makeup works in all sorts of ways, but removing it is basically powerless.