ChatGPT Images 2.5 adds sketch-led creation and more precise edits

ChatGPT Images 2.5 is rolling out with a stronger emphasis on revising an image after its first generation. OpenAI says the release improves detail, lighting and the preservation of subjects across edits, while reducing generation latency by up to 50% against Images 2.0.

The announcement covers ChatGPT, ChatGPT Work and Codex across desktop, mobile and web. The speed figure is OpenAI’s reported result; the time a user experiences will also depend on the request and chosen model.

Drawing the instruction

The new Sketch feature lets users supply a drawing as a reference. Image comments provide a way to direct attention to a particular region. Templates offer starting points for familiar formats, and a shared image can carry the prompt that produced it.

OpenAI’s developer announcement describes invoking Sketch with an @ mention. It also confirms the rollout across all tiers.

For someone making a sequence of product pictures, the useful test is consistency. A requested background change should leave the product’s shape, label and positioning intact. A second revision should preserve the first. Those are practical checks a creator can make with their own reference material.

Two API choices

The API release introduces GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst. OpenAI positions Flare as the general-purpose option, combining the quality improvements with faster generation. Sunburst is aimed at work needing more exacting edits, with longer generation times.

The release is useful to assess through a small, repeatable brief. Pick a recognisable subject, make several local edits, and compare the retained details after each step. That turns a broad claim about better editing into a result tied to the images someone actually needs to produce.