OpenAI’s ChatGPT Images gets faster, with sharper details and more refined edits
SiliconANGLE Mike Wheatley ● Covered by 4 sources
OpenAI updated ChatGPT Images with faster output and better edits. It now keeps details steadier, and you can sketch the image first.
Based on reporting by SiliconANGLE, Mike Wheatley — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI has pushed ChatGPT Images into version 2.5, and the pitch is simple: faster pictures, cleaner edits, and fewer mangled details. The company says the new model generates images with more natural lighting and richer textures, while also following instructions better across multiple turns in a conversation.
Speed is part of the story too. OpenAI says ChatGPT Images 2.5 can generate images up to 50% faster than before. That matters because the tool is meant for iterative work, not one-shot novelty. The model is supposed to hold onto the parts you didn’t ask it to change, even in busy scenes with several subjects and objects, so a targeted edit doesn’t slowly erode the whole picture.
The most interesting addition is Sketch. Users can draw directly in ChatGPT on a touchscreen and use that rough sketch as a reference for the generated image. OpenAI frames it as a starting point for a composition or visual direction, then lets the model turn that into something more polished. It also added templates for common formats like flyers and product photos, inline comments for targeted edits, and prompt sharing so other people can iterate on the same idea.
OpenAI is also splitting the model into two versions. GPT-Image-2.5 Flare is the default and is aimed at speed, with output that the company says is between two and four times as fast as GPT-Image-2 and better support for transparent backgrounds. GPT-Image-2.5 Sunburst goes the other way: slower, but more precise, with higher fidelity for use cases like product photos and marketing materials where details have to stay on brand.
The rollout is live in ChatGPT, ChatGPT Work and Codex on desktop, mobile and the web, and developers can tap both variants through separate APIs. OpenAI says the system includes prompt and output guardrails, plus C2PA metadata and invisible watermarking, so generated images can be identified and used against deepfakes and unauthorized reuse.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI image tools that actually matters: not bigger demos, but fewer broken edits. The industry keeps selling magic, while users mostly want the model to stop forgetting what they just told it. OpenAI is finally leaning into that boring, useful truth.
Read more about this at: SiliconANGLE