Condé Nast says a new multimodal search system reduced the average time its editorial teams spend finding archive video from 250 minutes to under two. The library contains more than 140,000 videos used across brands including Vogue, GQ, Vanity Fair and Wired.
The previous search relied mainly on titles and descriptions, forcing editors to scrub through files or depend on colleagues who remembered where footage lived. Working with the AWS Generative AI Innovation Center, Condé Nast built a system on Amazon Bedrock and OpenSearch Service that searches transcripts, visual elements and audio together. TwelveLabs’ Marengo model converts those signals into embeddings, numerical representations used to find conceptually related clips.
The system supports natural requests such as finding beginner yoga footage with calm backgrounds, even when those exact words are absent from a filename. The reported time reduction comes from Condé Nast’s own deployment and may not transfer directly to other archives. Its broader lesson is that video search improves when the index describes what viewers can see and hear, not only manually written metadata.