Meta Reveals Initial Output Results for Muse Video Model (3-minute read)

Meta is testing a beta version of the Muse video model with selected partners, and initial generated outputs have been released. This model was first teased in July alongside Muse image.

Initial test results from TestingCatalog showed that the model has state-of-the-art (SOTA) potential in terms of detail, world understanding, and temporal consistency. It currently generates 10-second videos, with output quality rated as very high.

Key features include:

• Native audio support: Audio generation capabilities are included, but there are still limitations in audio-video synchronization and the representation of physically accurate fast motion.
• Handling diverse scenarios: The model generated natural videos for various prompts, such as cyberpunk robots, shaky smartphone footage, logo-tech cats, and a tipping wine glass.

Although an official release date has not yet been confirmed, Meta has stated plans to first integrate the model into the Meta AI app and website. There is a high likelihood that the deployment scope will expand to Instagram, Facebook, and Meta's video editing app Edits in the future. This is part of Meta's strategy to build its own media models and integrate them into its creator and consumer product suites.