ChatGPT Images 2.5 Explained: Faster Generation, Better Edits, and the New Sketch Tool

OpenAI has launched ChatGPT Images 2.5, an upgraded image-generation system designed to create sharper pictures, preserve reference subjects more reliably, and make focused edits without changing everything else. Released on September 8, 2026, the update also adds new creative tools—including Sketch, templates, image comments, and shareable prompts.

OpenAI says people now create more than three billion images each week across ChatGPT Images and its image models in the API. The new release is therefore more than a visual-quality upgrade: it is an attempt to make AI image creation behave like an editable workflow instead of a one-shot prompt.

What changed in ChatGPT Images 2.5?

According to OpenAI’s launch announcement, the model focuses on four practical improvements: image fidelity, precise editing, consistency across multiple edits, and speed.

OpenAI says generation latency has been reduced by as much as 50% compared with Images 2.0. Actual waiting time will still depend on the complexity of a request, service demand, and the product being used, but faster generation matters because good visual work normally requires several attempts.

The company also says the new model produces more natural lighting and richer textures. When a user supplies a reference photo, distinctive features are more likely to remain recognizable after a change in setting, clothing, composition, or visual style.

Focused editing may be the most useful improvement

A common frustration with generative-image tools is unintended change. A request to replace a background may also alter a person’s face, shift the composition, rewrite text, or remove an object that was supposed to stay.

Images 2.5 is designed to isolate edits more accurately. A user can ask to update one product, color, label, or background while preserving the surrounding image. For businesses, that could make it easier to adapt an approved product photo into several campaign versions without rebuilding the design from the beginning.

OpenAI also highlights multi-turn consistency. In a longer conversation, previous edits should survive as new instructions are added. This is important for any workflow that moves through several stages—for example, changing a room’s wall color, adding furniture, correcting the lighting, and then producing a second aspect ratio.

The feature is still an AI system, not a guarantee of pixel-perfect control. Important commercial assets should be reviewed for unexpected changes, incorrect text, brand errors, and inaccurate product details before publication.

Sketch turns a rough drawing into a visual guide

The new Sketch feature lets a user draw directly in ChatGPT and use the drawing as a reference. A simple outline can communicate placement and proportion more clearly than a long written prompt.

Someone planning a room could draw where the sofa, window, and shelves should appear. A fashion designer could outline a garment’s silhouette. A marketer could sketch the location of a headline, product, and call-to-action. The user then describes the preferred style and lets the model turn the rough layout into a finished image.

OpenAI says Sketch can be started by typing @Sketch in ChatGPT. Feature placement may differ between desktop and mobile as the rollout continues.

Templates, comments, and prompt sharing

ChatGPT Images 2.5 also adds templates for common formats such as posters, merchandise, flyers, and product photos. Templates reduce the blank-canvas problem by asking for the information and design choices needed for a specific type of image.

Image comments offer a more visual way to request an edit. Instead of describing a location with words alone, a user can point to the part of the picture that needs attention and explain the change.

Prompt sharing is aimed at repeatable ideas. When sharing an image, a creator can choose to include the prompt so another person can try the same concept with different photos or details. Anyone reusing a prompt should still respect trademarks, copyrights, privacy, and the consent of people shown in reference material.

Two new API models: Flare and Sunburst

Developers receive two model choices through the Images API:

• GPT-Image-2.5 Flare is positioned as the default option for most applications, emphasizing quality and lower latency. OpenAI says it delivers higher quality than GPT-Image-2 with 50% lower latency.
• GPT-Image-2.5 Sunburst is aimed at premium creative work that benefits from tighter editing control, with longer generation times.

That distinction gives product teams a clearer tradeoff. A high-volume application may prioritize Flare for speed, while a polished campaign or product-image workflow may use Sunburst when precision matters more than turnaround time.

Availability

OpenAI says Images 2.5 is rolling out across all tiers of ChatGPT, ChatGPT Work, and Codex on desktop, mobile, and the web. The Flare and Sunburst models are also available through the API. A staged rollout means some users may see individual features later than others.

Safety and image provenance

More realistic image generation increases the risk of deceptive or non-consensual content. In its Images 2.5 system card, OpenAI acknowledges that increased realism can enable more convincing deepfakes if safeguards fail.

The company says it checks both prompts and generated images, continues to attach C2PA provenance metadata, and uses invisible watermarking to help identify content made with its tools. These measures can provide useful signals, but metadata may be removed when a file is edited or re-encoded. Viewers should still verify the source and context of consequential images.

How to get better results

A strong prompt should identify the subject, setting, composition, lighting, visual style, aspect ratio, and any text that must appear. When editing, it also helps to state what must remain unchanged.

For example, instead of asking for “a better product photo,” specify the product, camera angle, background, lighting, output format, and protected details. After each edit, inspect faces, hands, small objects, logos, and written text at full size.

Bottom line

ChatGPT Images 2.5 is most significant as an editing upgrade. Faster generation is welcome, but the ability to preserve subjects, maintain earlier changes, and target one part of an image could make the system more practical for everyday creative work.

The new Sketch and comment tools also reduce the need to describe every visual choice in words. For casual users, that means an easier creative process. For professional teams, it could mean faster prototypes and more reusable assets—provided that humans continue to check accuracy, permissions, and brand consistency.

Sources: OpenAI’s ChatGPT Images 2.5 announcement and the ChatGPT Images 2.5 system card, published September 8, 2026. Product-performance statements are attributed to OpenAI.