GPT Image 2 Prompt Guide: Composition, Text, and References

A useful GPT Image 2 prompt describes one visible scene: subject, arrangement, medium, lighting, and any exact words. In Pipe2 Image Generator, explicitly select GPT Image 2, paste your brief, choose an aspect ratio, and run. Auto may select another model.

Give the image a clear hierarchy

Start with the main subject, then place the supporting elements. “One orange desk lamp on the left, empty cream wall on the right” gives a clearer layout than “beautiful premium interior.” Choose one medium and a compatible light source rather than stacking contradictory styles.

For lettering, separate the scene from the copy. Quote each phrase and state which line should dominate. These examples are starting briefs, not tested outputs or promises of accuracy.

Two prompts to adapt

Product photograph — choose 16:9:

One orange enamel desk lamp on the left third of a pale oak desk. Eye-level product photograph, soft daylight from the left, cream plaster wall, gentle contact shadow. Keep the right half empty. Show the entire lamp and its base. No text, logos, or additional objects.

Check the silhouette, base, shadow, and empty space before changing the lighting. If the lamp is cropped, revise the framing instruction alone.

Exhibition poster — choose 3:4:

A minimalist exhibition poster with one orange desk lamp in the lower half against a cream background. At the top, render the exact headline “LIGHT STUDY” in large dark-blue sans-serif letters. Directly below, add the smaller line “Objects after dusk”. Keep both lines separate from the lamp. Render each phrase once, with no additional lettering or logos.

Inspect every letter and the hierarchy at the intended viewing size. If exact typography remains unreliable, add final text in an editor after generating the image.

Choose settings before revising the prompt

Select GPT Image 2 explicitly and start with Fast for an economical composition test. For this model, Fast maps to low quality, Auto to medium, and HD to high. Higher quality cannot repair a missing object or conflicting instructions. Check the current credit estimate before each run; revisions are separate generations.

The pipeline offers 1:1, 16:9, 9:16, 4:3, and 3:4. Set the destination shape in the form rather than relying on wording in the prompt. With GPT Image 2 selected, Pipe2 sends your written brief without its separate prompt-enhancement step.

Assign a role to each reference

GPT Image 2 accepts up to five image references here and no video references. Use only what the task needs: “Use image 1 for the lamp shape and image 2 for the palette; follow the layout in this prompt.” References guide appearance but do not guarantee exact geometry, identity, or lettering.

Keep the model, quality, ratio, and references fixed while changing one instruction. Review the composition at small size, then inspect edges, reflections, and text closely. Save the prompt with the accepted image. See the image generation guide for broader setup; if the correct photo already exists and needs a targeted change, use the image editing guide.