OpenAI Launches ChatGPT Images 2.5 with 50% Lower Latency and Precision Editing
Do Son 
OpenAI today unveiled Images 2.5, its new image-generation model, with a focus on improved image detail, editing precision, reference-photo consistency, and generation speed. OpenAI reports that ChatGPT Images and the GPT-Image family of models in the API together now generate more than 3 billion images each week.
Compared with Images 2.0, the new model reduces image-generation latency by up to 50%. It also produces more natural lighting and richer material textures. When working from a person, product, or other reference image, Images 2.5 is also more reliably faithful to the original subject’s appearance and distinguishing features.
Transparent-Background Generation Has Been Improved
OpenAI dedicated particular attention in Images 2.5 to transparent backgrounds. The model can now handle more complex layouts and images with transparent backgrounds. As a result, it delivers more reliable subject-and-background separation. This makes it well suited to producing logos, product assets, website elements, presentation illustrations, and images destined for further compositing in design software.
Early customer feedback highlighted by OpenAI specifically commends the improvement in transparent-background generation, noting its utility for brand assets, websites, and presentations. The Manus evaluation team, for instance, described it as an excellent fit for creative work. This includes brand assets, presentations, and websites.
Multi-Turn Image Editing Will Be More Stable
Images 2.5 places particular emphasis on strengthening localized-modification capabilities. When a user asks only to swap a product, background, or a piece of copy, the model endeavors to preserve the original subject, composition, and brand visual identity. Rather than regenerating the entire image, it focuses on requested changes. Earlier edits had a habit of disrupting a picture’s existing composition, and Images 2.5 shows meaningful progress in this regard.
OpenAI notes that, across multiple successive rounds of revision, the new model is more likely to retain edits already completed, reducing drift in subject appearance, compositional shift, and the gradual degradation of image quality as the edit count rises. This is especially significant for product photography, advertising assets, and design work that demands repeated refinement.
The API Offers Two New Image-Generation Models
For developers building image-generation workflows via the API, OpenAI is introducing GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst. Flare is the default model for most applications. It delivers higher-quality images than GPT-Image-2 at 50% lower latency. In addition, it is suited to product experiences, social content, visual search, and high-volume image generation.
Sunburst is built for premium visual workflows that benefit from tighter editing control, making it the better choice for professional designers generating high-quality advertising assets and product imagery. As one would expect, this tier commands a higher price.
Support Our Threat Intelligence
Find our tech and OS security coverage helpful? Support our work today and unlock a 100% ad-free reading experience!
Buy Me a Coffee
