OpenAI's DALL-E 3.6 Adds Animated GIF Generation

OpenAI pushed out Images 2.6 this week, adding animated GIF generation to a model that previously only handled static output, along with ChatGPT's annotation-based editing tools now available through the API. Developers also get a new dial for planning complexity, letting them trade processing depth for speed. OpenAI says the update runs 50% faster than the earlier GPT Image 2 release, though that's a launch claim rather than something verified under controlled conditions here.

The real question is whether Images 2.6 handles the fine details that separate a usable image from one that just looks good in a thumbnail. To test that, five hands-on prompts were run across generation, editing, and animation, starting with a scene built to expose exactly the kind of inconsistencies models tend to fumble: glass, condensation, snow, and warm interior light all occupying the same frame. A greenhouse on a snowy ridge at blue hour was the chosen subject, chosen specifically because it forces the model to reconcile materials and lighting that don't naturally cooperate. What follows is the prompt used and what Images 2.6 actually produced.