How to prompt ChatGPT Images 2.5
OpenAI · ChatGPT Images · Guidance v2, verified Sep 23, 2026 against 2 official sources
How to structure your prompt
Ask ChatGPT to create one image. Name the image type and intended use first (product photograph, ad, diagram, logo), then the subject, composition and aspect ratio, then visible details: materials, lighting, colors and medium. Say 'photorealistic' explicitly when that is the goal. For people, describe framing, gaze and interaction with objects. For complex requests use short labeled sections such as Scene, Subject, Details and Constraints. Keep requirements explicit without invented branding.
Length and syntax
Up to 2,000 characters. Longer prompts are not better if the extra words do not change what appears on screen.
Exact copy in quotation marks with its position and typography, and how many times it should appear; spell unusual words or brand names letter by letter when needed; add 'no extra text' when only that copy should appear. Short labeled sections (for example 'Constraints:') are allowed. Aspect ratio in words or chosen with the aspect ratio picker. A transparent background can be requested in words. No parameter flags.
Prompt skeleton
Create a [image type] for [purpose]. Scene: [setting]. Subject: [subject, framing, action]. Details: [materials, lighting, colors, medium]. Text: "[exact copy]" exactly once, [position], [typography]. Constraints: no extra text, no watermarks. [Aspect ratio in words].
What it’s best at
- legible text in images
- infographics, diagrams, slides and charts
- logos and cutouts with transparent backgrounds
- interface mockups
- real-world and historical context
- multi-panel comics and character consistency across turns
Known limitations
- editing an existing image is a separate workflow
- text, labels and fine details still require review
- selection-based edits are not always precise and can extend beyond the selected area
Avoid in your prompt
- unsupported command-line parameter flags
- invented lettering
- asking for an edit without supplying the image
- unrelated instructions mixed into text-heavy requests
Other image models
Summarised from the provider’s official documentation by Promptuned. Results vary by model version and settings. ChatGPT Images 2.5 is a trademark of its owner; Promptuned is not affiliated with OpenAI.