OpenAI Releases Prompt Guide for Image 2.6
November 10, 2026 · 2 min read
OpenAI has released an official prompt guide to help users effectively utilize Image 2.6. The guide emphasizes that users should specifically define the purpose, composition, and style to create accurate images, and clearly distinguish between elements to change and elements to maintain when modifying existing images.
Published on November 7 in the developer documentation, the guide introduces writing principles and practical examples for image generation and editing. In the developer API, Image 2.6 is referred to as GPT-Image-2.6. The documentation provides instructions on prompt writing and model usage through various production cases, including photographs, advertisements, infographics, logos, and GIF animations.
The basic principle is to first describe the purpose of the image and the desired result. OpenAI recommends separating the subject, composition, style, and mandatory conditions. For complex tasks, these can be divided into items such as scenes, subjects, details, and constraints. However, the guide notes that prompts do not need to be unconditionally long, as clear communication of requirements is more important than length.
When adding text, the guide suggests using quotation marks for exact phrases and specifying the location, font, and number of occurrences. Users are advised to manually check the spelling and readability of small or complex text after generation. For image modifications, users should specify both the target of change and the elements to keep. The guide recommends modifying one element at a time rather than making many changes at once.
Another method involves using a transparent image to mark the specific location for modification. For example, a user can mark a vase in a kitchen photo with a red circle and request a color change to blue while maintaining other objects and lighting. The model then removes the marking circle in the final result.
Image 2.6 is divided into two models: Flare, a lightweight model suitable for fast generation, and Sunburst, a base model for detailed depiction and precise editing. Additionally, the guide distinguishes between complexity and quality settings. Complexity determines how deeply the composition and element placement are planned, while quality affects visual details like textures and small text.
For GIF animations, the guide suggests instructing the model to connect the start and end of the motion and specifying camera movement. For logos, the SVG format can be used to maintain quality across different sizes. OpenAI also emphasizes verifying the accuracy of completed images, particularly for infographics and scientific diagrams, as visual design does not guarantee factual correctness.
Because the guide is based on the developer API, some model selections and advanced settings cannot be adjusted directly in the general ChatGPT interface. However, the core principles of specific descriptions and clear distinctions remain applicable to ChatGPT image generation.