OpenAI launches Images 2.5 with faster editing tools

OpenAI has launched Images 2.5, a new version of its image-generation tool, for ChatGPT users worldwide across all service tiers.

The update also extends to ChatGPT Work and Codex. Developers will get access to two related models through the API: GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst.

People and businesses already create more than 3 billion images each week across ChatGPT Images and the GPT-Image API models, giving the release a large installed user base from the outset.

According to OpenAI, the new model improves image quality with sharper fine details, more natural lighting and richer textures. Generation latency has also been reduced by up to 50% compared with Images 2.0.

Beyond image generation, the release adds several editing and collaboration features inside ChatGPT. These include Sketch, which lets users draw directly in the chat as a reference for a final image; new templates for common formats such as flyers and product photos; image comments for more targeted edits; and the option to share prompts alongside completed images.

Editing focus

OpenAI says Images 2.5 is better at preserving the key subject of a reference photo while changing settings, styles or compositions. The system is also more reliable at applying editing instructions to one part of an image without altering other details in the frame.

That approach is aimed at users making iterative changes over longer conversations. Earlier edits are more likely to remain consistent across multiple rounds of changes, reducing the need to rebuild an image from scratch.

For business users, that could matter in areas such as product imagery, marketing material and design mock-ups, where a single item, background or text element may need to change without affecting the rest of a composition.

New tools

Sketch is one of the most visible additions. Users can type @Sketch in ChatGPT and create a rough drawing to guide the final output. OpenAI says this can help with room layouts, clothing concepts and other visual ideas that are easier to draw than describe.

OpenAI is also introducing templates to help users start with a defined structure rather than a blank prompt. Examples include poster and merchandise formats, where users can add information, design direction and style preferences before generating an image.

Prompt sharing is intended to make images easier to reproduce or adapt. Users who share an image can now include the original prompt, allowing others to generate a related version using their own photos and details.

Developer access

For developers, the new API models split the offer between speed and detail. GPT-Image-2.5 Flare is positioned as the faster option, with the same broad improvements in quality and editing, while GPT-Image-2.5 Sunburst is designed for more detailed work with longer generation times.

OpenAI says the model is also better at handling complex visual instructions, more accurate when generating images that include real-world information, and more capable with layouts such as transparent backgrounds. It is also more likely to maintain a requested visual direction and composition as prompts become more specific.

Those changes reflect growing demand for image tools that support repeatable workflows rather than one-off experiments. As image generation becomes more widely used in workplaces, consistency, editing control and turnaround time are becoming more important than novelty alone.

Images 2.5 is now available across desktop, mobile and web for consumer users, workplace subscribers and developers through the API. Weekly image creation across OpenAI's platforms already exceeds 3 billion.