Emu Video

Emu Video is a generative video tool. Meta's research demo for text-to-video generation that first renders an image from a prompt, then animates it, producing 4 second clips at 512px and 16fps.

Emu Video is a research text-to-video generator from Meta's Fundamental AI Research group, published alongside a paper titled Factorizing Text-to-Video Generation by Explicit Image Conditioning. The site is built as a demo and a technical showcase rather than a commercial product: it presents the method, sample outputs and a link to try prompts in a browser. The core idea is a two stage pipeline. A first diffusion model renders a still image from the text prompt, then a second diffusion model animates that image while staying tied to the original prompt. Because generation is split this way, the team reports training efficient high quality video models without the deep cascade of stacked models used in earlier systems. Output is capped at 512 pixels, 4 seconds and 16 frames per second. In practice the demo is prompt driven. You type a scene description and the model returns a short clip; the gallery on the site shows the kind of prompt that works, mixing photoreal subjects like a bear on a ski trip or a person fishing in the ocean with stylised requests such as a bear in a graduation gown rendered anime style. Human raters compared the results against Make-a-Video, Imagen-Video, Align Your Latents, Reuse and Diffuse, Cog Video, Gen2 and Pika Labs, and the authors report wins on both quality and prompt faithfulness. It suits researchers, students and creative technologists who want to see image conditioned video diffusion in action or read the underlying paper, not production teams needing long clips, high resolution or commercial licensing. Anyone needing longer runtimes or editing controls should look to commercial generators, which typically trade this kind of published research transparency for output length and tooling.