ChatGPT Can Now Generate GIFs—but You May First Get a Policy-Violation Warning

OpenAI today released ChatGPT Images 2.6, a version update to its image-generation model. Core changes: support for generating GIF animations, saving reference images for reuse, two new API models, and a 50% increase in generation speed.

First, the official demo.

The GIF feature lets you start from a prompt or an image and directly generate a looping animation. Speed is adjustable, individual frames can be edited, and results can be shared. OpenAI gave an example: turning a brand mascot into an animated sticker, or turning a product image into an eye-catching looping short clip.

The reference-image feature is more practical. You can save reference images of yourself, your pet, or a character and call them up directly the next time you generate, without re-uploading.

On the developer side, there are two new models: GPT-Image-2.6 Flare and Sunburst. Flare focuses on speed, producing an image in 6 seconds at low-quality settings; Sunburst focuses on high quality and editing consistency. The API also begins supporting image markup and SVG export, making precise edits and downstream use easier.

In the official blog, one figure is given: ChatGPT, Codex, and the API generate 4 billion images per week, averaging more than 500 million per day. At that scale, a 50% speedup saves no small amount of time.

After the feature launched, users responded quickly. Some made meme images. For example, this user generated a looping animation of “Ram Altman.”

Someone else tried to get it to generate a stick-figure martial-arts fight, but received a notice: “We’re sorry, but the generated image may have violated our guardrails on violence. If you believe we made a mistake, please try again or edit your prompt.” Content moderation is still in place; if you want to use GIFs to make something edgy, you may first have to learn how to avoid keywords.

Another user demonstrated GIF generation speed from 9 frames to 24fps—basically instant results.

Chinese users also tried it out.

The GIF feature sounds like a throwback to the early internet, but OpenAI is aiming at social media’s sharing efficiency. Static images can only be looked at; GIFs can be replayed, forwarded, and turned into memes. For marketing and content creation, this is a lower-barrier tool for dynamic assets.

Another notable point is the complexity parameter in the API. You can control how much time the model spends planning before generation: use low complexity for quick edits, and raise it when the scene is complex and has many elements. This is more granular than simply adjusting quality parameters, making it easier for developers to balance speed and results.

Images 2.6 is now fully available in ChatGPT and Codex, supporting web, desktop, and mobile. The two API models are also open to all developers. The official pricing page has details.

As for whether the GIF feature will further flood the world with memes, and whether there will be more frequent tension between content moderation and creative expression, that’s another topic. At least for now, you can try generating a moving version of yourself first.