Emu Video

Emu Video is a research project from Meta AI that generates short videos from text prompts. It takes a description like "a hamster is driving an autorickshaw" and produces a 4-second, 512px video at 16 frames per second. The model is designed to be straightforward — it avoids the multi-stage cascades used by earlier systems by splitting the task into just two steps.

The first step generates a single image from the text prompt. The second step uses that image and the original prompt to produce a video. This factorized approach makes training more efficient and still delivers high quality results. In human evaluations, Emu Video outperformed several existing models, including Make-a-Video, Gen2, and Pika Labs, on both video quality and how faithfully the video matches the prompt. The generated examples on the site show a range of scenes — from a wildfire in a forest to two sloths playing chess in slow motion.

Emu Video is primarily a research demonstration. It's most useful for AI researchers and engineers interested in text-to-video generation, as well as content creators who want to experiment with generating short clips from text. The site includes links to the research paper and blog post, but there is no public API or commercial product. If you're curious about where AI video generation stands today, this is a solid example of the current state of the art.