OpenAI Releases Images 2.6 Adding GIF Generation and Reusable Photo References
Artificial intelligence research laboratory OpenAI introduced Images 2.6 on November 4, rolling out animated GIF generation alongside reusable reference photos. The updated generative image framework is deployed across ChatGPT, the Codex software development tool, and developer application programming interfaces. The functionality is accessible worldwide across desktop, mobile, and web clients for all subscription plans.
Company metrics show users assemble roughly 4 billion images weekly across these platforms, representing more than half a billion pictures daily.
Animation Features and Generative Velocity Improvements
Under the animation feature, users can describe intended physical movements or upload a still picture to produce looping animated GIFs. The platform allows users to adjust playback velocities directly within the conversational interface before downloading or distributing the completed animation. Modifications to animated outputs can be requested through text prompts, though the tool cannot directly edit pre-existing animated GIFs uploaded by users. The release expands creative workflows for marketing teams animating brand mascots, digital retailers presenting products on loop, and creators designing custom memes. Animators can also assemble kinetic storyboards to plan character choreography before initiating full video production sequences.
Persistent Visual Memory and Identity Consistency
A second key capability introduces persistent personal photo libraries that can be recalled across subsequent conversation sessions. Subscribers can upload an image, assign it a distinct identifier, and reference it in later conversations using an @ mention. The reference system allows prompts to request creative transformations of a saved subject without uploading the original photographic file again. The system maintains consistency across multi-turn edits, ensuring faces, clothing patterns, and product contours remain stable while backgrounds change. Images generated directly within ChatGPT are preserved under dedicated folders and remain available until the parent conversation is removed. OpenAI's GIF animation feature in Images 2.6 has no direct predecessor in the company's prior image releases, as every earlier system card evaluated static image generation, according to Webiano Digital. OpenAI's system evaluation determined that the GPT-Images-2.6 series with moderation checks performs on par with GPT Images-2.5, according to OpenAI Deployment Safety Hub. OpenAI's deprecations page removes gpt-image-1-mini, gpt-image-1.5, and chatgpt-image-latest from the API on December 1, 2026, according to Enjoy.dev.
Developer API Architectures and Provenance Controls
For commercial developers, the API introduces two specialized model options: GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst. The Flare variant is engineered for rapid generation, capable of producing visual assets in as little as six seconds on lower-fidelity settings. The Sunburst architecture emphasizes fine visual detail, maintaining structural and character fidelity across medium and high-resolution rendering tiers. The API additionally supports interactive canvas markup for localized edits alongside SVG export options for web, application, and print publishing. Generated still visual creations incorporate formal Content Credentials alongside an imperceptible SynthID digital watermark. GPT-Image-2.6 Flare and Sunburst models have identical published OpenAI token rates of $5.00 per million text input tokens, $8.00 per million image input tokens, and $30.00 per million image output tokens, according to OfoxAI. In API benchmarking of GPT Image 2.6 Sunburst on November 5, 2026, generating a 9-frame animated GIF took 32.6 seconds of latency and consumed 1,147 image output tokens, according to OmniaKey.
Background
SynthID watermarking robustness guarantees were established against PNG and JPEG files and have not been separately validated against GIF compression pipelines, according to Webiano Digital.