Technology
OpenAI Launches GPT-Image-2.5 With Faster Generation, Sketch Input, and Split API Models
By Today News Hours Editorial Desk
OpenAI has released GPT-Image-2.5, a new image generation model that cuts generation latency by up to 50% compared to its predecessor, Images 2.0, while improving editing precision and introducing a hand-drawn input feature called Sketch. The update is available to all ChatGPT users, including the free tier, as well as ChatGPT Work and Codex.
According to OpenAI, ChatGPT Images and the GPT-Image API together generate about 3 billion images per week, translating to roughly 430 million images per day.
One of the most notable improvements is the resolution of a long-standing issue with garbled Chinese characters appearing in generated images, which OpenAI says has now been eliminated. Demonstrations shared by the company show the model restoring old printed photos into high-definition digital images, generating realistic outfit try-ons, and correcting details such as messy bedding into neatly arranged scenes, all while maintaining strong consistency between reference and output images.
In terms of editing accuracy, OpenAI says Images 2.5 addresses a common flaw in image generation tools where requested local edits inadvertently altered other parts of an image. The new model is designed to modify only the specified area while preserving composition, lighting, and other elements, with improved stability in complex backgrounds and scenes involving multiple subjects. Users can also add comment-style annotations directly on images to mark areas for editing, a workflow similar to the design tool Figma. Additionally, the model retains editing consistency across multiple rounds of revisions within a single conversation, preventing image quality from deteriorating as edits accumulate.
Axultan Alimkulov, Head of Product at Higgsfield AI, noted that the model's ability to understand what should remain unchanged during edits stands out, as meaningful edits do not compromise the character, composition, or visual style of the original image.
On the product side, Images 2.5 introduces four new features. The Sketch function allows users to type