In 30 seconds
Bin Liu says his team found that structurally correct JSON or XML did not ensure an attractive video. They turned to HTML, CSS, and JavaScript as a representation that language models could use to express visual decisions.
- 1Structured output was not enoughBin says JSON and XML could be correct without yielding a visually compelling result.
- 2Move to web codeHis team turned to HTML, CSS, and JavaScript to express visual choices.
- 3Edit through conversationHe says the approach lets users direct an agent to revise videos.
Takeaways
- A structurally valid data format does not guarantee a good visual result.
- Bin’s team chose web code to express layout, styling, and behavior.
- He believes this representation makes conversational video editing more practical.
Read the edited text by chapter
From structured data to visual code
Bin explains his team’s approach to AI video editing.
Bin Liu says his team found that an agent could produce structurally correct JSON or XML without ensuring the resulting video looked good. They turned to HTML, CSS, and JavaScript as a way to express both information and visual styling. In his view, web code gives the model a useful representation for creating and revising videos through conversation. This is an account of his team’s experience, rather than a claim that a particular format guarantees visual quality.
More from this channel
46:40EN
17:50EN
15:03EN
33:09EN
13:05EN
1:06EN