OpenAI released ChatGPT Images 2.5 on Tuesday, an update to its image model that the company says produces “more natural lighting and richer textures,” keeps subjects from reference photos more recognizable and follows editing instructions more reliably over several rounds. Generation latency is down by up to 50% compared with Images 2.0, which came out in April.
The model is rolling out to ChatGPT, ChatGPT Work and Codex users on all tiers, on desktop, mobile and web, according to OpenAI’s announcement. OpenAI says people create more than 3 billion images a week across ChatGPT Images and its GPT-Image API models.
Sketch, comments and templates
The most visible addition is Sketch. Type “@Sketch” in ChatGPT and a drawing window opens; the model then turns the rough drawing into a finished image based on a text description. The Verge‘s Jay Peters said it turned a cat he drew with his mouse into a realistic photo, and he used the new comment tool, which lets users leave instructions on specific parts of an image, to change the cat’s eye color.
Templates for formats such as posters and product photos start with a basic prompt and then ask follow-up questions. In TechRadar‘s test, the product-photo template asked for an upload and then a style, such as Clean Studio or Lifestyle. TechRadar’s Graham Barlow also pointed to an Edit toolbar (Markup, Comment, Remove BG, Erase and Resize) that OpenAI appears to have rolled out during August without a formal announcement. Users can now share the prompt behind an image so others can make their own version.
Axios, which got early access, found the model did “a much better job preserving the likeness of people and pets” in tests that included restyling a photo of reporter Ina Fried’s cat. “The sketch tool and these templates are kind of an attempt to make people more engaged in the process of creation, rather than them being a passive force,” Adele Li, OpenAI’s product lead on images, told Axios.
Two API models, and Adobe Firefly
Developers get two new models. GPT-Image-2.5 Flare is the default for most applications and, OpenAI says, delivers higher-quality images than GPT-Image-2 at 50% lower latency. GPT-Image-2.5 Sunburst takes longer per image in exchange for tighter control, aimed at production campaign creative and polished product imagery.
Adobe has added the new models to Firefly, its creative AI studio. “We’re excited to expand that choice with OpenAI’s latest GPT‑Image‑2.5 models now in Firefly, bringing faster generation and resolution consistency that keep images sharp and photorealistic through every refinement,” Matt Chotin, senior director of product at Adobe, said in OpenAI’s announcement. Lucky Liao of Manus’s evaluation team said Flare ran at “two to four times the speed of GPT‑Image‑2” in its tests.
OpenAI said images made with its tools continue to carry C2PA metadata and invisible watermarking. The release follows OpenAI’s decision in March to shut down its Sora video app, which, the BBC reported, earned $1.4 million in global net in-app revenue compared with $1.9 billion for ChatGPT over the same period, citing Sensor Tower. OpenAI told the BBC at the time that ChatGPT’s image tools were not affected.
WebProNews is an iEntry Publication