OpenAI launches ChatGPT Images 2.5 with faster generation, Sketch and refined editing

The release is rolling out immediately to ChatGPT, ChatGPT Work, and Codex users on desktop, mobile, and web, according to OpenAI. The company also said the underlying image model is accessible to developers through two application programming interfaces.

The update comes about five months after ChatGPT Images 2.0, which brought 2K-resolution output, support for multiple aspect ratios, and the ability to use up-to-date web research. OpenAI said Images 2.5 is better at preserving the details of subjects in reference photos, including distinctive facial features, through multiple image generations and edits in a single conversation. Requested edits should no longer degrade details that the user did not ask to change, even in complex scenes with many objects or busy backgrounds.

A signature new feature, dubbed Sketch, lets users draw directly inside ChatGPT with a finger, mouse, or stylus and use that rough drawing as a visual reference for the image they want to generate. OpenAI said the model builds on the sketch to create a more polished image.

The update also adds templates for common image formats such as flyers and product photos. Users can now leave inline comments on parts of an image as instructions for targeted edits, and they can share prompts along with generated images so other people can iterate on an idea.

TechRadar, which said it tested the updated product for 24 hours, reported that an image-editing toolbar with tools including Erase, Remove BG, Resize, Markup, and Comment had been rolling out inside ChatGPT during August. The publication said the editing tools make it significantly easier to pinpoint specific areas for changes instead of typing several revised text prompts, and each edited version is saved separately so users can roll back to earlier images.

The Verge said it briefly tested Sketch by drawing a cat with a computer mouse and asking ChatGPT to turn the doodle into a realistic photo. According to The Verge, the feature followed the instructions and produced a realistic image, and a comment left on the image changed the cat's eye color from blue to green.

OpenAI is offering two models for developers. GPT-Image-2.5 Flare is the default for most applications, providing higher-quality images than GPT-Image-2 at 50 percent lower latency. GPT-Image-2.5 Sunburst sacrifices speed for greater precision, making it suited to premium creative and product photography workflows, the company said.

The company added that guardrails on prompts and outputs remain in place to prevent harmful content, and generated images use C2PA metadata and invisible watermarks so they can be identified as AI-generated. The Images 2.5 announcement follows last week's launch of GPT-6 Astra.