OpenAI has quietly turned ChatGPT into a GIF workshop. With the rollout of Images 2.6, the image generator the company deployed on 4 November 2026 across ChatGPT, Codex and its API, users can now produce animated GIFs from a text prompt or an uploaded photo and refine them before downloading. The capability is open to everyone, subscription or not.
- Free for all: GIF generation and editing arrive without a paywall; the subscription tier only changes how many images you can create.
- Conversational editing: Loop speed, content tweaks and frame-by-frame review are handled inside the chat interface.
- Faster engine: OpenAI says Images 2.6 renders images roughly 50% quicker than the previous version.
- Developer options: API users can pick between two model variants, Flare and Sunburst, depending on whether they prioritise speed, detail or coherence.
From a single sentence to a looping animation
The workflow is deliberately frictionless. You describe the GIF you want in the text box — a tortoiseshell cat mid-yawn, for instance — or start from one of your own images. OpenAI notes that even a rough sketch, such as a simple owl drawing, can serve as the seed for an animation. The model then builds the frames around that input.
Editing is where the feature earns its keep. A “1x” control in the corner lets you adjust the loop speed with a tap. Opening the image and typing a change into the “Describe edits” box modifies the content through the same conversational logic that has become standard across creative tools. A separate sheet view displays the frames one by one, with room for comments and annotations. Once satisfied, you download or share the result.
That approach mirrors a broader industry shift toward agent-driven editing inside familiar interfaces, rather than forcing users into dedicated design software. The GIF simply becomes another artefact you can produce without leaving the chat window.
Who gets access, and where the details run thin
Availability is broad by design. Images 2.6 is live for all users through ChatGPT and Codex, and the subscription level — or absence of one — determines only the image quota. OpenAI has not published specific allowance figures alongside the launch.
Developers integrating through the API face a choice between Flare and Sunburst, two variants tuned for different priorities: raw speed, fine detail or temporal coherence across frames. The distinction matters for anyone building animation pipelines at scale, since GIF output quality depends heavily on how consistently the model holds a subject together between frames.
Several practical questions remain unanswered. The announcement does not specify output formats, maximum file size or duration limits for generated GIFs, leaving users to discover the boundaries empirically. Those constraints will matter for anyone hoping to use the feature for anything beyond quick social sharing.
The scale behind the feature
Context here is considerable. OpenAI states that its tools generate more than 500 million images per day on average, though it does not break out how many of those are animated. Even a small slice of that volume would make ChatGPT a significant producer of GIFs globally.
The strategic logic is less about technical novelty than about default behaviour. Generating a GIF is not a breakthrough, but embedding creation and editing of animations inside a tool with hundreds of millions of users — free and without friction — nudges people toward ChatGPT for every quick visual task, whether that is a logo, a meme or a reaction GIF for a group chat. The goal is not to outpower dedicated design applications; it is to remove the reason to open one.
The same ease that makes the feature fun is what makes it problematic in other hands.
That convenience carries a familiar caveat. When anyone can produce realistic animations in seconds and share them without provenance markers or context, the GIF stops being purely humorous and enters the same contested territory as other synthetic media. Watermarking and labelling practices across AI video and image tools remain inconsistent, and the GIF format — widely stripped of metadata during re-sharing — is particularly poor at carrying attribution forward.
For now, the feature is a genuine quality-of-life upgrade for casual users and a useful prototyping shortcut for developers. Its longer-term significance will depend on whether OpenAI attaches durable provenance signals to animated output, and whether the unspecified format and size limits prove generous enough for real-world use. Until those details surface, the yawn-ing cat is the easy story; the metadata question is the one worth watching.