OpenAI released ChatGPT Images 2.5 today, delivering a 50 percent latency reduction compared to the April 2026 Images 2.0 model. The update also improves detail preservation, enables more reliable multi-turn editing, and introduces Sketch, a new drawing-based reference system that lets users sketch compositions directly in ChatGPT as a starting point for final images.
The model rolls out to all ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web. Two new API variants, GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst, are available to developers immediately.
What the 50% Speed Boost Includes
Image generation latency has dropped by up to 50 percent compared to Images 2.0. The model also produces more natural lighting, richer textures, and better preservation of subjects in reference photos. Distinctive facial features and other identifying details now carry through more reliably across multiple generations and edits within a single conversation.
How Editing Preserves Unrelated Details
Images 2.5 keeps details you have not asked to change intact, even across complex scenes with multiple subjects and busy backgrounds. Earlier models often degraded unrelated areas when users made targeted edits, but this version maintains stability across the entire image.
Over longer ChatGPT conversations, editing instructions remain consistent across multiple turns. Earlier changes stay stable, and each successive edit builds on prior work without quality loss, making iterative creative workflows far more practical.
Sketch: Drawing as a Reference Tool
Sketch lets you draw directly in ChatGPT as a visual reference for the final image. You provide rough layout, composition, or visual direction, and the model uses that as a starting point to generate a polished output. iPad and Apple Pencil users benefit most from this addition, since drawing on touchscreen is now a native part of the creative workflow.
Beyond Sketch, Images 2.5 adds templates for common formats like flyers and product photos, inline comments on images for targeted editing requests, and the ability to share prompts so others can iterate on your ideas with their own photos.
Two API Models: Speed Versus Precision
Developers choose between two variants. GPT-Image-2.5 Flare is the default: it delivers high-quality images at two to four times the speed of GPT-Image-2, with improved support for transparent backgrounds. It suits brand assets, presentations, and websites where turnaround matters.
GPT-Image-2.5 Sunburst trades speed for precision, offering extra detail fidelity for creative work that demands pixel-perfect consistency across product photography, marketing collateral, and design tools where brand accuracy is non-negotiable.
Safety and Watermarking
ChatGPT Images 2.5 enforces safeguards on prompts and generated images to prevent harmful outputs. The model applies C2PA metadata and invisible watermarking to all outputs, helping identify images created with OpenAI’s tools and combat unauthorized reuse or deepfakes.
