OpenAI Launches ChatGPT Images 2.5 With Sketch and Up to 50% Lower Latency

OpenAI has released ChatGPT Images 2.5, a new image generation model with sharper details, more precise editing, and faster generation. The update also adds Sketch, templates, tools for commenting directly on images, and prompt sharing to ChatGPT, according to the company.



The model produces more natural lighting and richer textures while doing a better job of preserving subjects from reference photos across different settings, styles, and compositions. Compared with Images 2.0, which arrived in April, generation latency has been reduced by up to 50%.



Those improvements extend to editing. Images 2.5 is better at changing only what the user requests while preserving the surrounding details, even in complex images and backgrounds.



Instructions also hold more reliably across longer conversations. Earlier changes are more likely to remain consistent as additional edits are made, without image quality degrading over time.



The new Sketch tool provides another way to begin an image. Typing "@Sketch" opens a drawing surface where users can outline a room layout, trace the contour of an outfit, or make a doodle before adding a description of the intended style and other details. ChatGPT then uses the drawing and description as references for the finished image.


Templates offer starting points for formats such as posters and merchandise. Once an image has been generated, comments can be placed directly on specific areas to request focused changes. Shared images can also include their original prompts, allowing other people to adapt the same idea with their own photos and instructions.


Developers receive the same model improvements through two new Images API options. GPT-Image-2.5 Flare is the default choice for most applications, producing higher-quality images than GPT-Image-2 at 50% lower latency. Potential uses include creator content, visual search, rapid prototyping, and high-volume generation. GPT-Image-2.5 Sunburst offers additional precision for detailed creative and editing work, with longer generation times.

The release builds on existing safeguards, including checks on prompts and images intended to prevent harmful outputs. OpenAI says it will continue using C2PA metadata and invisible watermarking to help identify images created with its tools.

Images 2.5 is rolling out today to ChatGPT, ChatGPT Work, and Codex users across all tiers on desktop, mobile, and web. GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst are also available through the Images API.