In early September 2026, OpenAI rolled out a major update to ChatGPT's image generation engine: ChatGPT Images 2.5. Building on Images 2.0, which launched back in April, this release promises faster generation, sharper detail, and much more precise editing tools. In short, if you heard about "ChatGPT 2.5," it's not the chat model that got the upgrade — it's the image side.
What Changed from Images 2.0 to 2.5?
According to OpenAI's release notes, the key improvements are:
- Faster generation: Image creation latency is down by up to 50% compared to Images 2.0.
- More natural lighting and texture: Images now come out with more realistic lighting and surface detail.
- Better reference consistency: A person's facial features or an object's distinctive traits hold up far better across multiple rounds of edits.
- More reliable text rendering: Posters, infographics, and UI mockups — anything text-heavy — show noticeably fewer spelling and letter-formation errors.
Standout New Features
Sketch — Users can draw a rough sketch directly in ChatGPT and feed it to the model as a reference using @Sketch. A room layout, an outfit idea, or a simple doodle can turn into a detailed image in seconds.
Targeted editing — No more describing an entire image from scratch. You can now drop a comment directly on the image to mark just the part you want changed — a chair's color, the background, a product — and edit only that region.
Templates — Ready-made templates (80s aesthetic, stickers, underwater scenes, anime, caricature, studio portraits, icon design, and more) give a quick starting point for anyone short on ideas.
For Developers: Two New API Models
On the API side, OpenAI introduced two new models: GPT-Image-2.5 Flare, built for speed, and GPT-Image-2.5 Sunburst, designed for work that needs tighter control. The split lets developers choose between speed and precision depending on the use case.
Safety
Every generated image continues to carry C2PA metadata and an invisible watermark. Safety filters against violent content, sexual content, and impersonation of real people were retrained on the 2.5 dataset.
Rollout: Who Gets It, and When
The update is rolling out today to ChatGPT Plus, Pro, Team, and Enterprise subscribers. Free-tier users are expected to get access in the coming weeks.
Where It Sits Competitively
The release lands at a moment when rivals like Midjourney v7, Flux 1.1, and Ideogram 2.0 are also shipping new versions focused on text handling and editing. OpenAI's edge is keeping image generation inside the same conversational flow as the LLM — you never have to leave the chat to design something.
Takeaway
ChatGPT Images 2.5 is a real step forward on speed and consistency. Reference-photo preservation, targeted editing, and the new Sketch feature in particular could meaningfully speed up work for both individual users and teams producing visual content. Expect more example outputs — and prompt ideas worth trying — to show up over the coming days.
