70 upvotes
New Benchmark Tests Agents That Write Code to Create Videos
What is VibeVideoBench: Evaluating Multimodal Agents on Code-Driven Video Generation?
VibeVideoBench measures how well multimodal AI agents can generate videos by writing code for each frame and syncing audio. It challenges models to produce exact formulas, accurate charts, rule-consistent games, or narration timed to on-screen events.
Who it is for
Researchers building AI video generators can use it to gauge progress and compare approaches.
How to try it
Open the benchmark page on Hugging Face