OpenAI has released ChatGPT Images 2.5, a model designed to generate visuals with improved instruction following and faster response times. The update claims better preservation of subjects from reference photos and supports multi-turn interactions. Two new API endpoints, gpt-image-2.5-sunburst and gpt-image-2.5-flare, are available for developers. OpenAI directs users to choose Sunburst when editing precision matters most, while Flare suits fast, high-quality everyday generation.
The practical impact lies in the ability to refine specific visual elements across multiple prompts without losing core subjects. This reduces the need for manual editing or starting from scratch when an initial output misses a detail. The release also signals a shift toward more integrated workflows where text and image generation interact more tightly within the ChatGPT interface.
* Sunburst prioritises editing precision over raw speed
* Flare targets fast, high-quality everyday generation
* Reference images help maintain subject consistency
