SAN FRANCISCO — A picture made in ChatGPT no longer has to stay still. OpenAI has launched Images 2.6 with GIF generation, letting users turn a written request or an existing image into a looping animation and then adjust its speed or individual frames.
The November 4 launch also adds reusable reference images and brings two new image models to developers. OpenAI says Images 2.6 is available globally in ChatGPT and Codex across subscription tiers on the web, desktop and mobile, with GPT-Image-2.6 Flare and Sunburst available through its API.
The practical shift is not just another way to make a meme. OpenAI is trying to shorten the distance between a visual idea, a revision and a reusable asset. But its headline speed claims describe particular generation settings, not a guarantee that every animation or editing job will finish in seconds.
In its Images 2.6 announcement, OpenAI says image generation is 50% faster than before. Its speed-focused Flare model can produce an image in as little as six seconds when quality is set to low. That qualification matters: the company does not present the figure as a universal completion time for GIFs.
GIF creation begins with a prompt or an image. The company’s official announcement thread says users can change animation speed, edit individual frames and share the result. A reusable reference library separately lets people keep pictures of themselves, pets or characters ready for future creations, instead of uploading the same material for every request.
That makes reference images distinct from the animation itself. One is material carried between projects; the other is the output assembled for a particular request. Saving a character reference can reduce repeated setup, but it is not a guarantee that every pose, outfit or brand detail will survive each subsequent edit unchanged.
OpenAI pitches the feature to people making reactions and memes, as well as marketers animating mascots or product photographs. It also describes animated storyboards as a way to explore a character’s movement before video production. Those examples position GIFs as a lightweight creative format, not evidence that the update replaces a full video-production workflow.
The company says even a simple sketch can serve as the starting point for an expressive character. Its launch page includes examples credited to early-access customers, including OpenArt, and describes Town, an AI work-assistant startup, experimenting with animated characters for workplace interactions. These are selected launch demonstrations, not independent tests of typical output quality.
Developers face a separate choice between the two API models. Flare emphasizes speed and improves layouts for posters, product imagery and professional visuals, according to OpenAI. Sunburst emphasizes higher-quality generation and greater consistency across edits when image quality is set to medium or high.
That distinction is important for anyone estimating what the update will do to a production workflow. A quick low-quality draft and a detailed final illustration are different jobs. OpenAI’s six-second example describes the former; it should not be combined with Sunburst’s quality claims as though both were one measured result.
The API also gains image markup, which was already available in ChatGPT. Teams can indicate the part of an image they want changed rather than relying entirely on a written description of its location. SVG exports give them another way to carry designs into websites, apps and print work.
A new complexity parameter introduces a second control alongside image quality. OpenAI describes quality as governing visual detail and fidelity, while complexity governs how much planning goes into meeting a request. A crowded scene with many constraints can call for more planning even when it is not intended as the final high-fidelity asset.
For most editing tasks, OpenAI recommends low complexity for faster changes, paired with the image-quality level the job requires. It suggests medium or high complexity for scenes with more elements or constraints. That separates the effort spent organizing the scene from the detail requested in the finished image.
Those controls could matter as much as the animation button for teams making repeated revisions. The cost in a design workflow is not only producing the first image, but also changing one part without repeatedly rebuilding everything else. OpenAI says Images 2.6 better preserves characters, products and brand details through editing rounds, although that remains a vendor claim rather than a result established by this report.
The scale figures are also OpenAI’s own. It says users now create four billion images a week across ChatGPT, Codex and the API. That counts generated images across services; it is not a count of distinct users, and the announcement does not break out how much of the activity involves GIF generation.
The company describes availability across subscription tiers, but that is not the same as promising identical usage allowances or unlimited generation. Its announcement directs API users to separate pricing information and does not establish one fixed price or turnaround time for every creative task.
For a creator trying the update, the open question is how much revision survives the move from a showcase example to an ordinary project. A looping character can be generated, slowed down and edited frame by frame. The harder test is whether the face, product label or outfit still matches after the next requested change.
