OPENAI has unveiled Images 2.5, the latest iteration of its image-generation model, promising better image detail, more precise editing, improved reference-photo consistency, and faster generation. The firm states that ChatGPT Images and the GPT-Image family together now produce over 3 billion images per week. Relative to Images 2.0, Images 2.5 cuts image-generation latency by up to 50% and delivers more natural lighting and richer material textures.
When starting from a reference image of a person, product, or other subject, the new model is also said to be more faithful to the original subject’s appearance and distinguishing features.
OpenAI has placed particular emphasis on transparent-background generation, noting improved handling of complex layouts and images with transparent backgrounds. This should yield more reliable subject-background separation, making it suitable for logos, product assets, website elements, presentations, and assets intended for further compositing in design software. Early customer feedback highlighted in the announcement points to this improvement as valuable for brand assets and marketing materials.
In addition, Images 2.5 aims to make multi-turn image editing more stable: across successive revisions, the model is reported to retain prior edits more reliably, reducing drift in subject appearance and composition, which is important for product photography and advertising workflows.
For developers, OpenAI introduces two API image-generation models: GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst. Flare is the default, offering higher-quality images at 50% lower latency and suitability for product experiences, social content, visual search, and high-volume generation. Sunburst targets premium, tightly controlled editing workflows for high-quality advertising assets and product imagery, at a higher price point. The report positions Flare for broad use and Sunburst for professional design applications. 9 September 2026.