How Condé Nast built multimodal video discovery with Amazon Bedrock
Amazon Web Services Mariah Miller
Condé Nast cut video search from 250 minutes to under 2. It now scans visuals, audio, and transcripts, so old clips aren’t buried in a 140,000-video archive.
Based on reporting by Amazon Web Services, Mariah Miller — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Condé Nast’s editorial teams had a simple but brutal problem: finding the right video took too long. On average, each content discovery task ate up 250 minutes, and the process depended on people manually scrubbing through a library of more than 140,000 videos. Titles and descriptions carried the load. That worked until it didn’t.
The company’s answer was to stop treating video like a pile of filenames. Working with the AWS Generative AI Innovation Center, Condé Nast built a multimodal discovery system on Amazon Bedrock and Amazon OpenSearch Service, with TwelveLabs’ Marengo model doing the embedding work. The system searches across transcripts, visual elements, and audio at once, then returns relevant clips with precise timestamps. Discovery time fell to under 2 minutes per task.
The architecture is split in two. An ingestion pipeline handles uploads, validation, chunking, and asynchronous embedding generation. A separate serving tier handles user queries, turning natural language into vector searches and pulling back results fast. That separation mattered during the initial backfill of the 140,000-video archive, because the expensive work of making video searchable could run without slowing down the search experience.
Amazon Bedrock gave the team access to the Marengo model through a single API, along with IAM, VPC isolation, and CloudTrail controls. Amazon OpenSearch Service handled low-latency k-NN search with multi-AZ replication and metadata filtering. The result was not just smarter search, but less dependence on tribal knowledge. When the person who “knows where everything is” is unavailable, the system still works.
Condé Nast measured the payoff in a May 2026 benchmarking workshop. It reported a 99.2 percent reduction in discovery time, more than 90 percent less manual review effort, and about $800,000 in estimated annual operational savings. Faster searches also mean faster responses to advertiser requests and sales opportunities, which is the real business point hiding inside all the infrastructure talk.
My take — AI-written commentary, not fact-checked reporting
This is the kind of AI story that actually earns its keep: not flashy generation, just less time wasted hunting through archives. The industry loves to talk about synthetic video; meanwhile, the money is often sitting in the old footage nobody can find. Multimodal search is the unglamorous cousin that does the job and goes home on time.
Read more about this at: Amazon Web Services