Skip to content

Emu3

One transformer for image, text and video via next-token prediction

Visit official site
Category AI models

What Emu3 is

The model is described as a multimodal system trained only by next-token prediction, unifying images, text and video into a single transformer.

What you can do with it

  • Generate images with flexible resolution and style
  • Produce video by predicting the next token in a sequence
  • Extend a clip naturally from a given video context

Model notes

  • Visual-language understanding gives coherent text replies
  • No diffusion or composite architecture is used
  • A discrete space carries all three modalities

Pros & cons

✓ What we like

  • Flexible resolutions and styles are supported
  • Image, text and video share one discrete space

! What to watch out for

  • Training relies only on next-token prediction
  • The listed flow starts by choosing a generation or prediction function

FAQ

How is it trained?

The material states it is trained only through next-token prediction.

What can it generate?

The material lists image generation, video generation and video prediction.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review