Overview
OpenAI on Tuesday announced ChatGPT Images 2.5, a new iteration of its image-creation technology that the company says reduces generation times by as much as 50% compared with Images 2.0 while improving visual quality and editing reliability. OpenAI reported that users produce more than 3 billion images each week across ChatGPT Images and the GPT-Image API models, and positioned 2.5 as an efficiency and capability upgrade for both end users and developers.
Visual quality and editing improvements
According to OpenAI, Images 2.5 yields more natural lighting and richer textures than prior versions. The model has been tuned to preserve the appearance of subjects in reference photos more faithfully and to follow multi-turn editing instructions with greater consistency. OpenAI highlights better handling of reference images so that subjects appear more recognizable while retaining distinctive characteristics across different settings and visual styles.
The company also said Images 2.5 is better at making targeted edits without unintentionally changing other aspects of the image, a capability that holds across complex subjects and backgrounds. The model is described as maintaining consistency through repeated edits during longer conversations, avoiding the degradation in image quality over time that can complicate iterative workflows.
New user features in ChatGPT
OpenAI added several tools for ChatGPT users to support image creation and editing. Sketch enables users to draw directly inside ChatGPT to use a hand-drawn outline as a reference for final images. Templates offer preset starting points for common formats, including flyers and product photography. Users can place comments directly on images to focus edits and can share prompts with others, streamlining collaborative or repeatable workflows. The update is available across desktop, mobile, and web to users of ChatGPT, ChatGPT Work, and Codex.
Developer-facing models
For API customers, OpenAI introduced two new GPT-Image-2.5 variants. GPT-Image-2.5 Flare is described as delivering the same quality, editing fidelity, and speed gains as Images 2.5. GPT-Image-2.5 Sunburst is positioned for tasks that require additional precision for detailed creative work but comes with longer generation times, reflecting a trade-off between accuracy and latency.
Capabilities around complex visuals
The 2.5 models are said to better comprehend complex visual instructions and real-world information, accommodating more intricate layouts, including transparent backgrounds, and more accurately reflecting specific visual styles. These enhancements aim to support use cases that require fine-grained control over composition and appearance.
Safety and provenance
OpenAI stated that Images 2.5 builds on its existing safeguards, continuing to apply checks on prompts and images to reduce the risk of harmful outputs. The company also continues to embed provenance information using C2PA metadata and invisible watermarking to identify images generated by its tools.
Availability
The update is available now to the stated ChatGPT product suites and to developers via the API models described above.