OpenAI Ships Images 2.5 With More Precise Editing
OpenAI has released Images 2.5 across ChatGPT and its API, combining faster generation with stronger reference fidelity and multi-turn editing.
OpenAI has released ChatGPT Images 2.5 across ChatGPT, ChatGPT Work and Codex, while adding two API models aimed at different creative workloads.
Faster generation, tighter edits
OpenAI says Images 2.5 improves reference-image fidelity, lighting, texture and adherence to complex visual instructions. It is also designed to preserve a subject, composition or brand treatment when users change only one element, such as a product, background or piece of copy.
The company reports up to a 50 percent reduction in generation latency compared with Images 2.0. In ChatGPT, a new Sketch feature lets users draw a rough idea and use it as an image reference. Templates for formats such as posters and product photos are intended to reduce the blank-page problem, while comments and shared prompts make iterative work easier to pass between people.
Two API paths
GPT-Image-2.5 Flare is the default API option for most applications, with OpenAI positioning it for creator tools, social content, visual search and high-volume generation. Sunburst is the higher-control option for detailed editing and production-oriented creative work, accepting longer generation times in exchange for more precision.
OpenAI says the models are already available through the API and the consumer rollout covers desktop, mobile and web. The company also says it continues to use prompt and image checks, C2PA metadata and invisible watermarking.
Why it matters
The release is important because image generation is moving from one-shot novelty toward revision-heavy production. The key capability is not simply sharper output; it is keeping the unchanged parts of an asset stable while a user makes several targeted edits. That is especially consequential for advertising, retail and design workflows, where inconsistency creates rework. The open question is whether the claimed gains survive demanding commercial use, where small errors in text, identity and layout remain expensive.