GPT Image 2
prompt guide.
Clear intent. Visible details.
A brief you can check against the result.
This is an independent guide. Official OpenAI documentation is linked beside the technical guidance. The example structure and workflow advice are our editorial suggestions, not a special model syntax.
Start with the image you need
Name the subject, purpose, arrangement, visual medium, and constraints. For a complicated scene, separate important requirements into readable sections. OpenAI’s image prompting guide describes shared fundamentals and includes a separate GPT Image 2 reference.
Where GPT Image 2.5 fits
OpenAI documents two GPT Image 2.5 variants: Flare, oriented toward speed, and Sunburst, oriented toward demanding image quality and precise editing. Their performance must be evaluated in your own workflow.
The optional GPT Image 2.5 preset here writes a brief for that family. GPT Image 2 stays the default. A writing target is separate from the operator-configured VISION_MODEL that analyzes the image. This site still returns text or JSON, never a generated image, and does not choose a Flare or Sunburst image-generation endpoint.
A useful structure, not a required format
Our suggested brief has eight fields: subject, scene, composition and viewpoint, lighting, style and materials, text and typography, colors, and aspect ratio. These fields help you review an idea. They are not official API parameter names or a requirement to use JSON.
Subject: A ceramic cup on pale stone. Composition: Eye-level, slightly left of center. Light: Soft window light from the left. Materials: Matte glaze and lightly textured stone. Text: No added text. Colors: Forest green, gray and chalk white. Aspect ratio: Landscape 3:2.
Length: choose clarity over a magic word count
Our editorial recommendation is to use enough detail to communicate the result, then remove competing instructions. A claimed exact ideal length is Unconfirmed here. The official GPT Image 2 cookbook provides worked examples; we do not treat a particular word count as a quality guarantee.
Text needs exact wording and placement
Quote the wording and explain where it belongs. OpenAI discusses spelling, positioning, and readability in its text prompting guidance. Inspect the generated lettering; prompt wording cannot guarantee an error-free result.
Editing: separate the change from the constraints
The official cookbook illustrates edits with explicitly preserved details. In your brief, distinguish the requested change from the areas you want to keep. Our tool writes an editing brief; it does not edit images.
Common brief problems to check
- Competing composition instructions or too many main subjects.
- Unquoted lettering that leaves the exact words unclear.
- Unnecessary camera claims inferred from appearance alone.
- Unclear edit boundaries that omit what must stay the same.
- Assuming the original prompt or hidden settings can be recovered from pixels.
This checklist is editorial advice. Compare your own results instead of treating a writing preset as a guarantee.
Where the model ends and the tool begins
See the official GPT Image 2 model reference for model information. This site calls a separate, configured vision model to draft text. Selecting GPT Image 2 here changes the writing preset; it does not generate an image.
Sources and scope
Reviewed October 26, 2026. The current official prompting page also discusses other image models; use its GPT Image 2 reference for specific details. We make no unsupported promises about latency, exact reproduction, hidden negative-prompt syntax, or account access.