admin The upgrade, announced on September 8, is being made available across ChatGPT, ChatGPT Work and Codex on desktop, mobile and web. OpenAI said the model produces sharper details, more natural lighting and richer textures, while preserving people, pets and other subjects from reference photographs more consistently through successive edits.
A central change is Sketch, which lets users draw a rough visual reference rather than relying entirely on text. On mobile, users can type the @ symbol, select Sketch, draw a composition and then add written instructions describing the intended image. The feature is designed for cases where positioning, shape or layout is easier to indicate visually than explain in a prompt.
ChatGPT Images 2.5 also introduces image comments that allow users to indicate where a change should be made and describe the adjustment directly on the picture. OpenAI says the underlying model is better at changing only the requested element while keeping surrounding subjects, composition and visual treatment intact. The company said this should reduce a common problem in generative editing, where correcting one detail can unexpectedly alter other parts of an image.
The model is also intended to maintain consistency across longer editing sessions. OpenAI said earlier changes are more likely to survive subsequent instructions and that image quality should degrade less as users refine the same asset over multiple turns. Independent early testing has likewise highlighted stronger preservation of individual likenesses and more targeted edits, although performance can still vary with the complexity of the image and instruction.
OpenAI says the model also handles more complex instructions and infographic layouts with greater accuracy, including text placement and transparent backgrounds. For reference-led work, it is designed to retain facial features, clothing, objects and composition more faithfully when users change settings or styles, an improvement aimed at reducing the need to regenerate an image from scratch after each revision.
Generation speed is another major part of the update. OpenAI says latency has been reduced by up to 50 per cent against Images 2.0, a claim that refers to maximum improvement rather than a guarantee that every image will be produced in half the time. Faster turnaround is particularly relevant to iterative workflows in which users create several versions before settling on a final design.
Templates provide another starting point for common formats such as flyers and product photographs. Users can select a template and customise its details, with ChatGPT able to ask follow-up questions to refine the request. OpenAI’s release notes say templates are not yet available in ChatGPT Work, even though the Images 2.5 model itself is rolling out there.
Users can also share the prompt behind a generated image so that others can adapt the idea with their own photographs, wording or details. OpenAI said the added controls are meant to make the image workflow more interactive, shifting some tasks from purely descriptive prompting towards direct visual guidance and focused correction.
For developers, OpenAI has released two API models. GPT-Image-2.5 Flare is positioned as the faster option, carrying the core improvements in image quality, editing and speed, while GPT-Image-2.5 Sunburst is designed for tasks requiring greater precision and accepts longer generation times. The models also support more complex layouts, more accurate real-world information in images and improved transparent-background generation, according to the company.
The launch comes as image generation has become one of ChatGPT’s most heavily used creative functions. OpenAI says users create more than three billion images each week across ChatGPT Images and GPT-Image models accessed through its API, giving improvements in editing reliability and latency significance for both consumer and production use.
