Skip to content

OpenAI

Sketches become editing input in ChatGPT Images 2.5

Written by an AI model from the linked sources and checked against them by a second model. How we work

All about: OpenAI models and products →
Image: openai.com

OpenAI on September 8 released ChatGPT Images 2.5, an update to the image generator inside ChatGPT. OpenAI’s announcement says the update turns ideas, sketches and reference photos into images that are “more personalized” and “polished” and that “better reflect your ideas.”

It replaces ChatGPT Images 2.0, the gpt-image-2 release from April 21, which took first place across every LM Arena image generation category with a 242-point Elo lead over the nearest competitor. Its successor came out less than five months later.

The biggest change is how you edit. Axios reported that the update adds tools for building generations and edits on a sketch and for marking on the image where a change should go. Until now, fixing one corner of an image meant describing that corner in another prompt. Now you can point at it. Axios also reported that OpenAI says Images 2.5 generates images up to 50 percent faster than the previous version.

In Axios’s early testing, the updated engine kept the likeness of people and pets better and handled more detailed edits.

Anthropic

Local MCP servers run outside Claude’s VM sandbox

The author of a Claude desktop teardown wired a small local MCP server into the app just to see where it would run. It executed on the Mac under Claude, outside the Linux VM that runs agent code, and read a file never shared with the guest.

No comments yet. Be the first.

Join the conversation

Sign in to leave a comment. It appears after moderation.