22.【Generative AI Experiment】Verifying with Temple Architecture

Comparing 10 Prompt Directive Variations in Image Generation

tags:

🎯 Introduction

In the previous article, we used “Kyoto landscapes” as the theme and examined
how differences in prompt directive words affect image-generation results.

This article is a continuation of that experiment.
Here, we fix the theme to “Temple Architecture” and apply the same comparative method.

Unlike landscapes, architecture has clearly defined:

As a result, prompt directives act much more like explicit design inputs.

🎯 This experiment focuses not on mood, but on how structural directives take effect.

🧠 Why Choose “Temple Architecture”?

Temple architecture is an excellent subject for image-generation experiments.

In short:

The difference between vague and precise instructions becomes visually obvious

If the previous “landscape” experiment leaned toward atmosphere,
this one is firmly structure- and design-oriented.

🔬 Experimental Conditions (Fixed Rules)

The following conditions are fixed throughout the experiment:

🔬 Variables are minimized so that the effective change can be clearly identified.

⚙️ Base Prompt (Common)

The following prompt is used as the common base,
and only the differences are modified in each experiment.

Japanese temple architecture,
wooden structure,
clean composition,
natural lighting,
no text, no logo, no watermark

⚙️ Priority is given to observing differences, not overall image quality.

🔬 “Temple Architecture” Fixed — 10-Pattern Experiment Table

No. Change Type Prompt Difference Main Visual Change Observation Focus
1 Noun photograph of Japanese temple architecture Photo-like realism Attraction to real temples
2 Noun illustration of Japanese temple architecture Iconic rendering Organized composition and color
3 Abstraction abstract visualization of temple architecture Simplified forms Limits of structural retention
4 Specificity realistic wooden temple structure Structural emphasis Clear columns and beams
5 Emotion calm and quiet temple architecture Static feeling Reduced saturation and staging
6 Emotion dramatic temple architecture Theatrical Emphasized light and shadow
7 Era Edo period Japanese temple architecture Style fixation Era overrides realism
8 Contemporary modern reinterpretation of temple architecture Reinterpretation Changes in materials and form
9 Viewpoint aerial view of temple complex Bird’s-eye layout Placement and axiality
10 Viewpoint interior view of temple hall Interior space Column layout and depth

🔬 Generated Image Comparison (10 Patterns)

Temple architecture 10 patterns experiment

Figure 1: Ten generated results with temple architecture fixed,
changing only directive words in the prompt.

🧠 Observations and Findings

① In Architecture, “Structural Words” Have Top Priority

Words such as:

constrain output far more strongly than in landscape themes.

👉 In architectural prompts,
noun precision directly equals structural precision.

② Even When Abstracted, Structure Does Not Fully Collapse

Even with abstract visualization specified, elements such as:

are preserved to a certain degree.

🧠 Temple architecture is a structurally robust subject.

③ Emotion Words Act on the “Decorative Layer”

Rather than altering the building itself,
these words mainly affect light, sky, and surrounding elements.

⚠️ ④ Era Specification Completely Overrides Style

Terms like Edo period instantly fix:

⚠️ Extremely powerful, but prone to overriding other instructions.

⑤ Viewpoint Changes “How the Structure Is Understood”

👉 Choose the viewpoint based on what you want the viewer to understand.

⚙️ Practical Implications

✅ Summary

By progressing from landscape → architecture,
one gains a deeper understanding of how image-generation prompts operate.

End.