A good GPTImage2 prompt starts by explaining the job of the image. The model needs more than a subject; it needs a scene, composition, style direction, and constraints.
Prompt structure
Use this order:
- Goal: what the image is for, such as a product page, ad concept, blog cover, or diagram.
- Subject: the main object, person, UI, scene, or idea.
- Composition: framing, camera angle, aspect ratio, negative space, and focal point.
- Style: lighting, color, realism, texture, medium, and mood.
- Constraints: what must stay unchanged and what should be avoided.
Example
Create a clean product hero image for a lightweight desk lamp. Show the lamp on a white studio surface with soft shadows, a front three-quarter angle, realistic metal texture, and room for headline text on the left. Avoid extra props, distorted geometry, visible logos, and unreadable text.
Revision checklist
If the first result is close but not finished, revise one dimension at a time: composition, lighting, subject detail, background, or output format. Changing everything at once makes it harder to learn what improved the image.