Visual prompt method: from image intent to a reusable prompt

1. Start with image intent

State what the image must communicate, who it is for and which visual evidence matters most. Define the subject as well as action, distance, scale and causality between subjects.

For example, “turn a scientific principle into an observable miniature” gives a model more useful structure than “science, miniature, beautiful, 3D.”

2. Organize the prompt in six layers

Subject and relationships

Define the lead subject and how elements connect, act or align.

Composition and space

Set viewpoint, shot size, depth, negative space, movement and aspect ratio.

Materials and craft

Name materials and their observable surface qualities.

Light and color

Assign light direction, softness, time, dominant color and accents.

Camera and layout

Specify lens, depth of field, perspective and editorial hierarchy.

Constraints

Exclude unwanted text, watermarks, broken structures, extra objects and style drift.

3. Preserve a universal core, then adapt

ChatGPT Images and Gemini respond well to explicit natural-language relationships. Midjourney often benefits from compressed visual phrases and parameters. FLUX and SDXL commonly separate positive direction from exclusions. Keep a model-neutral core because versions and syntax change.

4. Review against visible evidence

Check subject accuracy, spatial continuity, material credibility, unwanted text and omitted relationships. For historical, scientific, product or cultural subjects, separate visual invention from factual claims.

Method questions