OpenAI Images 2.6 can now generate animated GIFs

What exactly arrived in Images 2.6

The key change in Images 2.6 is straightforward: OpenAI has added full animated GIF generation to ChatGPT and its API. The model can create a short animation from a text prompt or animate an uploaded image. OpenAI's official Images 2.6 announcement and API guide also state that the model is available in Codex.

The API offers two variants. GPT-Image-2.6 Flare is designed for rapid iteration, while GPT-Image-2.6 Sunburst targets more detailed imagery and a steadier visual result. This is a sensible split: testing motion drafts and producing a final asset have very different cost profiles.

The API workflow is fairly transparent. You set GIF as the output format, and the response arrives as an encoded file that must be decoded and saved with the appropriate extension. The frame_count and frame_duration_ms parameters control the number of frames and the duration of each frame.

OpenAI's official example uses nine frames at 200 ms each, producing a 1.8-second loop. At publication time, the CLI reference lists a range of 2 to 16 frames, with 9 as the default. For seamless animation, OpenAI recommends describing the starting pose, the object's return to that pose, and the camera behavior separately.

The current limitations are noticeable. GIF generation does not support streaming, partial images, compression settings, or prompt_rewriter. GIFs also cannot be produced through the editing API, although ChatGPT users can upload a finished image and ask for it to be animated.

According to the announcement, Images 2.6 generates images 50% faster than its predecessor. Flare can return a low-quality result in six seconds. These are OpenAI figures; the available materials do not include an independent assessment of loop smoothness or frame-to-frame consistency.

Why GIF is more than just another format here

The update genuinely shortens the path from a static image to a short animation. Creators gain a way to animate characters, mascots, and product shots, while developers of multimodal applications can assemble this output through the same image API.

I would focus less on the impact of a single successful example and more on geometric stability between frames, alignment of the first and last states, camera motion, and reproducibility after edits. Those are usually the points where generative animation stops feeling coherent.

There is also a separate safety layer. In its system card, OpenAI notes that motion can create sexualized or harmful meaning that is absent from an individual frame, so GIFs receive additional checks.

The feature looks useful rather than decorative. Its real level will be determined not by the ability to export a GIF, but by how rarely motion breaks the original scene between the first and final frame.

We previously reviewed Seedance 2, a video model with native 2K output and audio synchronization. That analysis helps compare OpenAI Images 2.6 GIF animation with the capabilities of specialized video-generation tools.