Image Generation Prompting Guide (38-min read)

gpt-image-2 is presented as the default for production image workflows. It excels at high-quality rendering, editing stability, text rendering, complex composition, and identity preservation, and for speed-first tasks, quality: low is often sufficient.

As of April 21, 2026, the model lineup is organized as follows.

• gpt-image-2: supports low, medium, high, with input_fidelity disabled
• gpt-image-1.5, gpt-image-1, gpt-image-1-mini: support input_fidelity for low, high, using 1024x1024, 1024x1536, 1536x1024, auto
• The recommended strategy is to move new work to gpt-image-2, keeping existing models only for migration validation or legacy compatibility

The size constraints for gpt-image-2 are also clearly defined. The maximum edge must be under 3840px, both edges must be multiples of 16, the aspect ratio must be 3:1 or less, and total pixels must be between 655,360 and 8,294,400. Beyond 2560x1440 (2K), results become more inconsistent and are considered experimental territory; the document's 4K/UHD example of 3840x2160 must be reduced to a value like 3824x2144 when the max-edge rule is strictly applied.

Rather than listing lengthy prompts, it's more effective to clearly state structure and constraints. Write in the order background/scene → subject → key details → constraints, and state the final use case upfront—such as advertising, UI mock, or infographic—to match the mode. Formats can include minimal prompts, descriptive paragraphs, JSON-like structures, instruction-style, or tag-based approaches, but in production environments, easily-scannable templates are preferred.

At a detailed level, materials, shapes, lighting, composition, a person's pose and gaze, and elements to preserve should all be specified. When photorealism is needed, directly including photorealistic is effective, and text should be specified with quotes or ALL CAPS, along with font, color, and placement. For multi-image input, each image should be distinguished with numbers and descriptions, clearly stating how to combine them, and rather than cramming too much into one attempt, iterating with small edits makes debugging easier.

The hands-on example begins with an infographic. This is suitable for explanatory materials for students, executives, customers, or the general public, posters, labeled diagrams, timelines, and visual wiki assets, and high quality is recommended for small text and dense layouts. The setup example continues with API client initialization, creating output_images/, adding a save helper, and an infographic prompt explaining the internal flow of a coffee machine.