OpenAI continues on its ‘code red’ warpath with new image generation model

OpenAI is rolling retired a caller version of ChatGPT Images that promises amended instruction-following, much precise editing, and up to 4x faster representation procreation speeds.

The caller model, dubbed GPT-Image-1.5, is disposable starting Tuesday to each ChatGPT users and via the API. It’s the latest escalation successful the contention with Google’s Gemini aft OpenAI CEO Sam Altman past period declared a “code red” successful a leaked interior memo. The memo elaborate OpenAI’s plans to regain its presumption arsenic the AI person aft Google had begun to instrumentality marketplace stock pursuing the release of Gemini 3, its latest flagship model, and Nano Banana Pro, the newest mentation of Google’s viral representation generator – some of which person topped the LMArena leaderboard crossed aggregate benchmarks. 

Google maintains its pb adjacent aft OpenAI responded to its occurrence past week with the launch of GPT-5.2, pitching it arsenic its astir precocious exemplary yet for developers and mundane nonrecreational use. OpenAI had reportedly been readying to merchandise a caller representation generator successful aboriginal January, accelerating those plans with this week’s announcement. Its past representation exemplary merchandise was GPT-Image-1 successful April. 

GPT-Image-1.5 arrives arsenic representation and video generators beforehand beyond prototypes and summation much production-ready capabilities. Like Nano Banana Pro, ChatGPT Images offers post-production features, providing much granular edit controls to support ocular consistency, similar facial likeness, lighting, composition, and colour code crossed edits. 

The aforesaid representation depicted supra with antithetic edits to showcase GPT-Image-1.5’s improved instruction-following. Image Credits:OpenAI

Most GenAI representation tools are atrocious astatine iteration, truthful this would beryllium a immense measurement up. Asked for a circumstantial change, similar ‘adjust the facial expression’ oregon ‘make lighting colder,’ models volition often reinterpret the full image, starring to a deficiency of consistency. 

The update isn’t conscionable astir caller features. ChatGPT Images volition besides present beryllium accessible via a dedicated introduction constituent successful the ChatGPT sidebar that works “more similar a originative studio,” Fidji Simo, OpenAI’s CEO of applications, wrote successful a blog station Tuesday. 

“The caller representation viewing and editing screens marque it easier to make images that lucifer your imaginativeness oregon get inspiration from trending prompts and preset filters,” Simo wrote.

On apical of the caller representation generator, OpenAI is introducing caller ways to amended the ChatGPT acquisition with much ocular elements. The program is to marque hunt queries show much visuals with wide sources, which could beryllium adjuvant for tasks similar converting measurements oregon checking sports scores, per Simo. 

“When you’re creating, you should beryllium capable to spot and signifier the happening you’re making. When visuals archer a communicative amended than words alone, ChatGPT should see them,” Simo wrote. “When you request a speedy reply oregon the adjacent measurement lives successful different tool, it should beryllium close there. As we bash this, we tin support closing the region betwixt what’s successful your caput and your quality to bring it to life.”