OpenAI’s new GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst models are now available in Adobe Firefly, extending the company’s latest image-generation technology beyond ChatGPT. The release accompanies the rollout of ChatGPT Images 2.5, which OpenAI says can cut latency by as much as 50% compared with Images 2.0 while delivering more detailed output and more precise edits.
Flare is positioned as the default GPT-Image-2.5 model for most applications, with OpenAI promising 50% lower latency than GPT-Image-2. Sunburst prioritizes more deliberate editing control over speed and is aimed at production assets and product imagery. Both models are available separately through the API.
Within ChatGPT, Images 2.5 adds tools intended to make image work more iterative. Sketch lets users provide a drawing as a reference, while Templates offers starting points for content such as flyers and product photos. Users can also leave comments directly on an image to identify the areas that need changing.
The updated model is also designed to preserve people and objects from reference photos more consistently when they are placed in new scenes or rendered in different styles. Images 2.5 is rolling out on desktop, mobile and the web for ChatGPT, ChatGPT Work and Codex users across access tiers.
Another sharing feature allows a creator to attach the original prompt to an image. Recipients can reuse that prompt with their own photos and details, turning a completed image prompt into a reusable template rather than sharing only the generated result.
The integration puts OpenAI’s models inside Adobe’s broader multi-model Firefly strategy. Adobe has also introduced more than 70 tools directly in ChatGPT, as the two companies expand their presence across each other’s creative workflows.
OpenAI says users generate more than 3 billion images each week through ChatGPT Images and GPT-Image API models. The company says its generated-image systems retain C2PA metadata and invisible watermarking for content identification, though the practical reach of those signals will depend on support across the platforms where images are shared.
