ChatGPT Images 2.5 is already here: Faster, more realistic, and with a new feature we weren’t expecting

Five months after ChatGPT Images 2.0, OpenAI is once again taking the next step in image generation with ChatGPT Images 2.5. The new version arrives after the unveiling of GPT-6 Astra and improves a tool that was already delivering impressive results. According to OpenAI, every week we create more than 3 billion images across ChatGPT Images and the API’s GPT-Image models. Now these images will be more precise and more realistic.

More natural images and much more reliable edits

Images 2.5 can generate more natural lighting, richer textures and, above all, better preserve the people and objects already present in reference photos. Latency drops by up to 50% compared with Images 2.0, so responses are substantially faster while the system manages to preserve changes better across several edits. We can ask for a specific modification and consistently preserve composition, subject, and style while we evaluate the changes.

ChatGPT Download

Sketch turns a quick drawing into a complete image

On top of that, Images 2.5 adds an interesting feature: Sketch lets us draw directly inside ChatGPT and use that sketch as a guide to create the resulting image. We can sketch a piece of furniture in a room, outline the cut or size of a garment, or simply make a doodle and let Images 2.5 complete the scene.

The new system also supports templates, comments on images, and lets us share our prompts.

Images 2.5 arrives on ChatGPT, ChatGPT Work, and Codex on desktop, mobile, and web starting today. OpenAI is introducing two models for developers, GPT-Image-2.5 Flare and Sunburst: the first prioritizes speed and volume, and the second prioritizes accuracy.

Clearly, a very interesting evolution for turning our ideas into images ever closer to what we really have in mind.