[Experiment] Improving a Beach Bikini Prompt in 8 Steps | Where Does Quality Actually Change?

Conclusion

• Step 4 (pose + shooting style specification) was the biggest turning point — adding hand on hip and swimwear photography dramatically changed image quality
• Camera model names have little effect in z-image-turbo — adding shot on Canon EOS R5 produced no visible difference
• Word order and element selection matter more than fine-tuning emphasis syntax — the difference between :1.2 and :1.3 is hard to see in z-image-turbo’s 8 steps
• Motion elements are effective as a finishing touch — dynamic elements like hair blowing in sea breeze and waves crashing work well as the final push
• wet skin is one of the most effective keywords for swimwear photos — the glossy, water-droplet-covered skin dramatically improves realism

“Good prompt writing tips” are covered in many articles, but do those techniques actually work? I verified this by starting from a minimal prompt and adding one element at a time to see how the image changes.

All images were generated with z-image-turbo.

Experiment Rules

• Theme: “Japanese woman in a bikini on the beach”
• Incrementally add elements from Step 0 (minimal prompt) through Step 7 (complete version)
• Record what was added and its effect at each step
• Negative prompts unused across all steps (verifying only the positive prompt’s effects)

Step 0: Minimal Prompt (3 words)

NSFW - クリックで表示

Result: An image was generated, but full of problems.

• Bikini color/design is random (came out as striped)
• Background is rocky, lacks “beach” feel
• Composition is awkward (full body but stiff standing pose)
• Age is unclear

Takeaway: Even with the bare minimum words an image comes out, but every element is determined randomly.

Step 1: Add Age, Pose, and Action

NSFW - クリックで表示

Changes: a Japanese woman in her 20s specifies age, standing on a beach makes the action explicit

Result:

• Bikini changed to solid navy (random element change)
• Still stiff upright pose and rigid expression
• Background slightly improved but still rocky

Takeaway: Specifying age has limited visual impact. Composition is specified (standing) but lacks specificity.

Step 2: Specify the Environment and Color

NSFW - クリックで表示

Changes: beautiful, black bikini (color specification), tropical beach (specific environment), sunny day

Result:

• tropical addition causes palm trees to appear — background dramatically improved
• Bikini is now the specified black
• beautiful addition makes the face slightly more refined
• Composition is still simple front-facing standing

Takeaway: Environment modifiers (tropical) have a dramatic effect on the background. Color specifications are reliably reflected.

Step 3: Skin Texture and Technical Settings

NSFW - クリックで表示

Changes: string bikini (specific design), wet skin (skin texture), bright sunlight (lighting), 85mm lens (camera setting)

Result:

• wet skin produces water droplets and gloss on skin — gravure feel jumps dramatically
• 85mm lens naturally blurs the background
• A hand gesture is added to the pose (touching hair)
• string bikini results in a string bikini

Takeaway: wet skin is one of the most effective keywords for swimwear photos. Camera lens specifications also affect bokeh.

Step 4: Explicit Pose and Composition

NSFW - クリックで表示

Changes: confident pose with hand on hip (pose specification), ocean waves in background (background detail), glistening in sunlight (more specific gloss), swimwear photography (shooting style)

Result:

• Pose dramatically improved — a confident pose with hand on hip
• Composition becomes standing at the water’s edge with added dynamism
• swimwear photography unifies the overall gravure feel
• Closest to professional swimwear gravure so far

Takeaway: Specific pose specification (hand on hip) directly affects composition. Genre specifications like swimwear photography instantly bring the overall atmosphere together. This is the biggest turning point.

Step 5: Adding Camera Model

NSFW - クリックで表示

Changes: shot on Canon EOS R5 (camera model), shallow depth of field (explicit bokeh specification)

Result:

• Almost no difference from Step 4
• Camera model name has no major impact on image quality
• shallow depth of field is redundant since 85mm lens already produces bokeh

Takeaway: Camera model specifications have little effect in z-image-turbo. When 85mm lens is already specified, shallow depth of field is redundant. Better to use the limited token budget on more effective elements.

Step 6: Adding Emphasis Syntax

NSFW - クリックで表示

Changes: swimwear photography emphasized to 1.3x, wet skin to 1.2x, shallow depth of field to 1.2x

Result:

• Subtle difference from Step 5
• Palm trees intrude too much into the frame, blocking the subject
• Emphasis syntax effects are limited in z-image-turbo (8 steps) (see experiment results)

Takeaway: In z-image-turbo, word order and element selection matter more than fine-tuning emphasis syntax. Adding appropriate elements like in Step 4 is more effective than relying on emphasis syntax.

Step 7: Final Version — All Elements Optimized

NSFW - クリックで表示

Changes: stunning (stronger than beautiful), toned athletic body (body type specification), turquoise ocean water (specific ocean color), golden sunlight (golden hour), hair blowing in sea breeze (motion), waves crashing (wave spray), alluring (direction of expression), high fashion magazine quality (quality instruction)

Result:

• Wave spray added — dynamism dramatically improved
• hair blowing in sea breeze makes the hair flow, adding motion to the image
• golden sunlight shifts the lighting to warm tones
• The most complete result of all steps

Takeaway: In the final stage, motion elements like waves crashing and hair blowing are effective. Changing the color temperature of light with golden sunlight significantly changes the atmosphere.

All-Step Comparison

Most Effective Prompt Elements from This Experiment

High Impact (Must Include)

1. Shooting style (swimwear photography, gravure) — determines overall atmosphere
2. Specific pose (hand on hip, arms crossed) — directly affects composition
3. Skin texture (wet skin glistening) — improves realism
4. Environment modifiers (tropical, turquoise) — dominates the background
5. Motion elements (hair blowing, waves crashing) — adds dynamism

Low Impact (Can Skip)

1. Camera model names (shot on Canon EOS R5) — little effect in z-image-turbo
2. Fine-tuning emphasis syntax (:1.2 vs :1.3) — hard to see difference in 8 steps
3. shallow depth of field (redundant with 85mm lens)

Most Efficient Prompt (Step 4 base + Step 7 motion elements)

This is the best cost-effective prompt structure found in this experiment:

Extracting only the good parts from Steps 4–7 and removing ineffective elements (camera model, redundant emphasis syntax). Kept to approximately 40 tokens.

Summary

The most important findings from prompt improvement:

1. Step 4 was the biggest turning point — specifying pose and shooting style dramatically changes image quality
2. “More is not better” — redundant elements like in Steps 5–6 have no effect
3. Motion elements are effective as a finishing touch — hair blowing, waves crashing as the final push
4. In z-image-turbo, word order and element selection > emphasis syntax — adding appropriate elements is more effective than fine-tuning

Related Articles

⚠ 関連記事が見つかりません: /en/tips/prompt-basics

⚠ 関連記事が見つかりません: /en/tips/prompt-design-thinking

⚠ 関連記事が見つかりません: /en/prompts/pose-prompts

⚠ 関連記事が見つかりません: /en/prompts/lighting-prompts

⚠ 関連記事が見つかりません: /en/prompts/gravure-prompts