Link CopyTwitterLinkedinWhatsappChatGPT Images 2.5 Adds Sketch. The Real Upgrade Is Editing ControlChatGPT Images 2.5 is OpenAI’s 8 September 2026 image-generation update. It adds Sketch, which lets users draw directly in ChatGPT as a visual reference, plus comment-based edits, templates and stronger multi-turn editing. OpenAI also launched GPT-Image-2.5 Flare and Sunburst for developers. Sketch is useful but it is not a technical first: Adobe Firefly already offered sketch-guided generation and drawn markup, while Microsoft leads the independent image-editing leaderboard used in this analysis as of 9 September. OpenAI’s new 2.5 models have not yet been independently ranked, so the company’s state-of-the-art claim remains unverified on launch day.OpenAI has given ChatGPT a pencil, but the more consequential part of ChatGPT Images 2.5 is what happens after the first picture appears. Type @Sketch and a rough drawing can become a finished image. Place a comment on one part of an image and ChatGPT can be told to change that element rather than reinterpret the whole frame. OpenAI says the model is also better at preserving faces, objects, composition and earlier edits across a longer creative session, while generation latency has fallen by as much as 50 per cent against Images 2.0. Those changes attack the least glamorous and most expensive failure mode in generative imaging: a model that produces a convincing picture, then destroys three correct details while fixing the fourth. The competitive question is not whether sketch-to-image is new. It is not. Adobe already lets creators guide Firefly with structural sketches and drawn markup, Midjourney has reference-led regional editing, and Microsoft’s latest model tops a major independent editing arena in the 9 September snapshot. The question is whether OpenAI can combine generation, reference fidelity, targeted editing and conversational iteration into a workflow that is easier than the specialist tools around it, without asking users to surrender control each time they press send.Key highlightsOpenAI launched ChatGPT Images 2.5 on 8 September 2026 with Sketch, comment-based editing, templates and improved multi-turn consistency.OpenAI says image generation is up to 50 per cent faster than Images 2.0, but ChatGPT Images 2.5 itself has not yet appeared on the current Artificial Analysis image leaderboards.Sketch is new to ChatGPT, not to generative imaging. Adobe Firefly already supports sketch-guided structure references and AI Markup with drawing and targeted instructions.OpenAI’s predecessor, GPT Image 2, still leads Artificial Analysis for text-to-image, while Microsoft MAI-Image-2.6 leads the editing leaderboard.The listed API token rates for Flare and Sunburst are twice those of GPT Image 2, so lower latency does not automatically mean lower generation cost.OpenAI has disclosed broad training-data categories and its safety stack, but not the architecture, parameter count or detailed model design behind Images 2.5.What actually launched with ChatGPT Images 2.5ChatGPT Images 2.5 is both a model update and a product-interface update, and separating those two matters. The model-level changes are the claimed improvements in detail, lighting, texture, reference fidelity, targeted edits, complex layouts and multi-turn consistency. The product-level changes are Sketch, templates, image comments and prompt sharing. Sketch is not a separate image model or a new API endpoint. It is a ChatGPT interface that turns a drawing into visual conditioning for the image-generation system.About The AuthorAt heart, I am a storyteller drawn to the watershed moments that bend the technology landscape. I braid narrative with data, humanise statistics, and trace the arc from first spark to world-changing impact. My reportage, features and reviews are witty, sardonic, visual and vivid, using anecdote to illuminate rather than eviscerate.