OpenAI released ChatGPT Images 2.5, its new image generation model. It renders sharper detail, more natural lighting and better likenesses from reference photos. Edits now change only the requested element and hold up across many rounds. Generation is faster than the previous version. OpenAI also added Sketch, which turns a drawing made in ChatGPT into a finished image. Templates cover common formats such as posters, flyers and merchandise. Users can place comments on an image and share the prompt behind a result. Two API models arrived alongside it: Flare for most uses and Sunburst for slower, more precise work. Images carry C2PA metadata and invisible watermarks. The model is live for ChatGPT, ChatGPT Work and Codex users on all tiers.
What changed
Images 2.0 was slower and less reliable at keeping unedited parts of an image intact.
What it unlocks
Drawing a rough sketch in ChatGPT and having it turned into a finished image.
- latency down up to 50% vs Images 2.0
- over 3 billion images created weekly
- 2 new API models
Sources