Turn visual direction into words and specifications

Give instructions through composition, color, light, shape and purpose.

Suggested reading time: about 8 minutes. This is an editorial production guide. Generation, performance and pricing comparison experiments have not been run.

Read first: Use AI to propose candidates under constraints

What you will learn

• Break abstract preferences into observable elements.
• Set priorities in a prompt.
• Assign elements requiring precision to a later production stage.

Describe more than a style name

“High quality” and “stylish” do not tell you what to change. Observe whether contours are rounded or sharp, forms simple or intricate, the palette limited, shadows soft, and the subject distinct from the background in brightness. When collecting examples you like, explain them in columns for composition, color, texture, light and purpose, rather than organizing the description around creator names.

A basic prompt structure

Purpose → subject → composition → style → light and color → background → constraints is a useful order for review. For example: “For a service guidance card. A round white mascot holds a small flag. Full body, centered, facing forward. Matte toy-like texture. Soft light from the upper left. Transparent background. Keep the fingers separate from the flag and do not draw text.” The order itself is not magic; the structure helps reduce omissions and contradictions.

Work backward from the destination

A circular icon, a wide hero image and a vertical post need different subject placement, even when they use the same image. Adding text in the implementation or a design tool makes revision easier. If you ask generation to handle precise text placement or typography, inspect spelling and layout independently. Reserving empty space during generation and checking readability on the final screen are separate tasks.

Work and responsibilities

• Art director: defines the visual rules.
• Graphic designer: handles placement, typography and color.
• Prompt operator: translates the rules into model inputs.

From input to output

Input	Workflow	Output
The aspect ratio of the destination screen.; Descriptions of the reference shapes, colors and lighting.; The area reserved for text.	1. Prepare a frame at the final display size.; 2. Sketch only three elements: the subject, empty space and light–dark balance.; 3. Write the prompt using a basic structure.; 4. Set the composition first, then refine texture.; 5. Place the result in the actual screen and adjust cropping and text readability.	A reusable art-direction note.; Images and crop plans for each use case.

Quality checks

• The main subject remains recognizable when scaled down.
• Faces and hands are not clipped at the screen edges.
• Decorations do not overpower body text or buttons.

Diagnose failures

Symptom	Possible cause	Correction
The image looks good, but there is no room for text.	Empty-space requirements were added afterward.	Extend the background or rebuild the composition.
Each service has an unrelated visual style.	There are no shared rules for color, shape and light.	Separate fixed elements from the elements that vary by service.

Exercise: Adapt one theme to three uses

Status: not run (not-run). This is a plan with completion criteria. No artifacts, duration, cost, performance or success rate have been recorded.

Write variation instructions for using one mascot as a circular icon, an empty-state illustration and a wide landing-page image.

Deliverable: A shared specification and variations for each use.

Completion criterion: Each proposal clearly states what stays fixed and what changes.

When you run it, separate predictions from results and record conditions and evidence using the experiment notebook.

Related practical templates

• Friendly rounded 3D mascot (Catalog candidates retained for review)
• Mascot for empty and completion states (Catalog candidates retained for review)
• Landing-page background with a character (Catalog candidates retained for review)

Technical references and evidence scope

Technical references were checked on 2026-10-03. They support the technical concepts. The exercises, workflows and evaluation criteria are editorial proposals, rather than experiments performed by the source authors. Recheck changing APIs, prices and publication rules when using them.

• OpenAI Image generation guide — Image generation and editing, output settings, and limits on consistency and precise placement.

Read the production flow

1. 1Destination and dimensions
2. 2Composition and copy space
3. 3Prompt constraints
4. 4Candidate
5. 5Crop and readability check

This is an original conceptual process diagram; arrows show the review sequence, not a claim about an undisclosed model architecture.