The OpenAI developer documentation "Image generation" page provides a centralized explanation of image generation and editing capabilities: developers can directly generate, edit, or generate variations through the Image API (variants are only available in DALL· E 2 supported), and "image_generation" can also be used as a built-in tool in the Responses API for multi-round conversational iterative restructuring and more flexible input methods (supporting file ID as image input). The document also gives a model comparison, saying that gpt-image-1.5 is the currently the main GPT Image series model, and prompts DALL· E 2 and DALL· E 3 has been marked as deprecated and is scheduled to end support on May 12, 2026.
In terms of output control, the document lists adjustable parameters such as size, quality, format, compression, and background transparency, and provides an auto option for the model to be automatically selected as prompted. The default output is base64 encoded image data; Formats are png, jpeg, webp, with jpeg generally being better for latency reduction. Transparent background needs to be set to transparent and is only available under PNG and WebP, with medium or high quality. Cost and latency are directly proportional to the tokens required to generate: larger size and higher quality significantly increase output tokens; If you use partial_images for streaming generation, you will be charged additional for each "partial image". The documentation also suggests that some organizations may need to complete API organization validation before using GPT Image models.
Frequently Asked
Questions Q: What is the main content of this "Image generation" document?
A: The document explains how to generate and edit images using OpenAI API, and compares the applicable scenarios and capabilities of Image API and Responses API.
Q: Which models are recommended for image generation first?
A: The document recommends using the GPT Image series, emphasizing gpt-image-1.5 as a higher-quality option and gpt-image-1-mini for lower cost.
Q：DALL· Can E 2/3 continue to be used and when will it be stopped?
A: The document is marked with DALL· E 2 and DALL· E 3 is deprecated and is scheduled to end support on May 12, 2026.
Q: How to generate transparent background images and which formats are supported?
A: Set background to transparent; Transparent only supports PNG and WebP, and medium to high quality settings are generally more stable.
Q: What are the main factors affected by the cost of image generation?
A: Fees are related to tokens, and the larger the size and higher the quality, the more expensive it is. Using editing or hi-fi input increases the input tokens, and partial_images also incurs additional output tokens fees.
