OpenAI Launches ChatGPT Images 2.5 With 2x Faster Generation, Hand-Drawn Sketches, and New API Models

OpenAI has introduced a major update to its suite of graphics tools: the ChatGPT Images 2.5 model. Developers sped up generation times by up to 50%, improved detail retention during repetitive edits, and added the ability to guide the AI using simple hand-drawn sketches.
Hand-drawn sketches, templates, and edit stability
A major pain point in previous versions was “degradation shift,” where attempting to modify a single detail would cause the AI to redraw faces, lighting, and the overall style of the image. According to OpenAI’s official announcement, Images 2.5 maintains key object characteristics across extensive sequential editing.
Several practical features have been added to the ChatGPT interface:
- Sketch tool: allows users to draw a rough sketch directly over the chat, turning it into a compositional framework for the upcoming generation;
- Pointed comments: users can highlight specific areas on a generated image with a mouse and write text feedback for individual details;
- Format presets: cropping and layout presets for creating product listings, merch, and promotional posters.
Journalists at Axios who received early access to the update noted improved accuracy when rendering text on logos and correct generation of transparent-background objects.
ChatGPT Images 2.5—faster, sharper, smarter, with better tools for creating whatever you can dream of.
— OpenAI (@OpenAI) September 8, 2026
– Faster image generation to keep your ideas flowing
– Improved fidelity for more natural, recognizable images
– Consistent details across multiple edits
– Comment-based… pic.twitter.com/SjHITKJUSV
Two API models: Flare and Sunburst
As part of the commercial API, the generator has been split into two separate models:
- GPT-Image-2.5 Flare is designed for bulk generation, social media content, and rapid prototyping. The model delivers results roughly twice as fast as the previous GPT-Image-2;
- GPT-Image-2.5 Sunburst is aimed at studio work and commercial printing. The model takes longer to generate imagery, but it adheres more precisely to strict prompt requirements during complex multi-layer retouching.
Access to ChatGPT Images 2.5 opens on September 8 for users across all subscription tiers (including free accounts, Plus, Pro, Business, and Enterprise), as well as in the Work and Codex environments. All generated files are marked with C2PA digital signatures and an invisible watermark to trace image provenance.