OpenAI’s Images 2.6 adds native GIF creation and faster generation

OpenAI’s Images 2.6 adds native GIF creation and faster generation

OpenAI has begun rolling out Images 2.6, a new image model that brings native animated GIF creation to ChatGPT for the first time. The update reaches all ChatGPT and Codex users on web, desktop and mobile, and also arrives on the company’s developer platform through two API variants. OpenAI says the model generates images 50% faster than its predecessor.

  • Key Point: ChatGPT can now build an animated GIF from a text prompt or an uploaded photo, then adjust playback speed and edit individual frames.
  • Key Point: OpenAI reports that more than 4 billion images are created each week across ChatGPT, Codex and the API, up from 3 billion in September.
  • Key Point: Developers get GPT-Image-2.6 Flare for low-latency work and GPT-Image-2.6 Sunburst for higher quality and stronger consistency across edits.
  • Key Point: The API gains image markup, SVG export and a new complexity parameter that controls how much planning the model performs.

From still frames to looping animation

Until this release, ChatGPT could not produce a GIF in a single step. Users had to generate separate frames and stitch them together in another application. Images 2.6 collapses that workflow: the model produces a set of related variations of the same subject and assembles them into a flipbook-style loop. Improved subject consistency keeps a character recognizable from one frame to the next, a requirement that earlier models struggled to meet.

OpenAI’s demonstration uses the prompt “Can you make a gif of a cat dancing?” to show the result, alongside controls for changing playback speed and stepping through frames one at a time. The company positions the feature for marketers animating a brand mascot, creators building reactions and memes for social media, and teams using animated storyboards to test character movement before committing to video production. Even a rough sketch, OpenAI notes, can become an expressive animated character for videos or games.

Town, a startup building an AI work assistant, used Images 2.6 through the API to explore how animated characters could add personality to everyday workplace interactions.

Two API models and tighter editing controls

The developer release splits into two variants, continuing the strategy OpenAI introduced with Images 2.5. Flare targets speed: with quality set to low, it can return an image in as little as six seconds, and OpenAI’s published benchmark shows generation time falling from 7.08 seconds on the previous model to 4.84 seconds. Flare also improves layout quality for posters, product imagery and other professional visuals. Sunburst, aimed at detail-heavy work, delivers higher-quality output and greater consistency across successive edits when quality is set to medium or high.

Editing receives several upgrades. Image markup, already available inside ChatGPT, comes to the API so teams can mark precisely where a change should land rather than describing it in prose. SVG export turns designs into vector assets ready for websites, apps and print. A new complexity parameter lets developers decide how much planning the model invests in a request, with lower settings suited to quick edits; OpenAI recommends low complexity paired with whatever image quality a workflow demands for most editing tasks.

What the numbers say about adoption

The weekly image figure of more than 4 billion, roughly half a billion per day, covers all image generation across ChatGPT, Codex and the API, not GIFs alone. It marks a gain of about a billion images per week since September, when OpenAI last reported 3 billion.

We’re introducing Images 2.6 in ChatGPT, Codex, and the API, bringing your ideas to life faster.

The rollout follows a busy stretch for OpenAI’s image tools. Images 2.0, released in April, was the company’s first image model designed to reason before drawing. The GIF capability also lands months after OpenAI shut down Sora, its AI video application, in March.

Several details remain unspecified in the release, including maximum GIF duration, frame count, dimensions and file size, along with rate limits for the two API models. Developers estimating production costs will also need current API pricing, since the announcement includes no new price sheet. GIF generation carries no separate paywall, though existing account-level usage caps still apply, and Codex now shares usage limits with ChatGPT.

The practical effect is consolidation: animation, reference reuse, targeted editing and vector export now sit inside one pipeline instead of being stitched together from separate tools. For teams building interfaces that need images on demand, sub-five-second generation makes synchronous creation more viable than pregenerated assets or background jobs. Frame-based animation still has limits, though, and short loops and stylized motion will hold up better than complex physical interaction or scenes with several moving subjects. Workflows with strict brand or geometry requirements will keep needing review and retries, which means the faster model shortens the loop but does not remove the human check at the end of it.