MicrosoftRecently launched aMoraThe video generation project aims to reproduceSoraMora uses a multi-AI agent framework to integrate multipleCutting EdgeVisual AI agents working to achieve the general video generation capabilities demonstrated by Sora.
Paper address: https://arxiv.org/html/2403.13248v1
Mora’s key features include:
- Convert text to video: Mora can generate videos related to the text content input. This means that users only need to provide a piece of text, and Mora can generate corresponding video content for it.
- Convert images to videos based on text conditions: In addition to directly converting text to video, Mora is also able to convert images to videos based on text conditions. This allows users to use existing image resources and combine them with text descriptions to generate creative videos.
- Extending generated videos: Mora has the ability to extend generated videos. It can extend and modify existing videos according to user needs to meet different application scenarios.
- Perform video-to-video editing: Mora also supports video-to-video editing, where users can splice and edit multiple video clips to achieve richer video effects.
- Concatenate videos and: Mora can concatenate multiple video clips to form a complete video story. This allows users to combine multiple independent video clips into a complete video work.
- Simulating the digital world: Mora also has the ability to simulate the digital world and can generate video content with specific themes and styles based on user needs.
Although Mora's performance on these tasks is close to Sora's, there is still a certain performance gap between the two when evaluated overall. However, Mora's successful experimental results provide a new direction for the development of future video generation technology, which is to achieve it through the collaborative work of multiple AI agents. Currently, Mora supports the generation of 12-second videos with a resolution of 1024*576.
Microsoft's Mora project demonstrates a new multi-AI intelligent agent framework that integrates multipleCutting EdgeThe visual AI agent has achieved a replica of Sora's general video generation capabilities. This project is expected to lead the development direction of future video generation technology and bring richer and more diverse video content to users.