News / AI & Data
Condé Nast cuts video search time from 250 minutes to 2 minutes thanks to multimodal AI Published on 29 September 2026 by Christ-loisele (2 min read)
Condé Nast has deployed an artificial intelligence solution based on Amazon Bedrock and OpenSearch to index over 140,000 videos. Editorial teams at Vogue, GQ, and Vanity Fair now save hours of work per task thanks to semantic and visual search capabilities.
Video (Vimeo) A radical productivity boost for media
Before adopting this solution, Condé Nast’s editorial teams spent an average of 250 minutes per task sifting through a library of over 140,000 videos, relying solely on titles and descriptions, according to the AWS Machine Learning blog . With the new system, this time has been reduced to less than 2 minutes , thanks to multimodal indexing combining visuals, audio, and transcriptions.
Condé Nast’s editorial teams now save hours of work per task thanks to semantic and visual search that tolerates typos and focuses on intent.
Image: Two-plane architecture: an asynchronous ingestion and indexing pipeline and a synchronous query and serving tier (Amazon Web Services, official image) An innovative technical architecture
The solution relies on two technical pillars: an asynchronous ingestion pipeline and a synchronous query and serving tier , as explained in the AWS blog . Videos, stored in Amazon S3, are segmented and their multimodal embeddings generated via the TwelveLabs Marengo model on Amazon Bedrock. These embeddings are then indexed in Amazon OpenSearch Service to enable vector similarity search.
The pipeline uses Amazon ECS with AWS Fargate and an Auto Scaling group to process segments in parallel, optimizing performance.
An intuitive and precise search
Users can now formulate natural language queries, such as ‘beginner yoga content,’ and retrieve relevant clips with precise timestamps, according to Dataforcee Digital . The solution tolerates typos and focuses on intent rather than exact keyword matches. It also enables image-based search: a user can upload a reference photo to find visually similar content in the archive.
What this changes here
For Beninese or West African media and administrations managing large video archives, such a solution could automate content search and drastically reduce production times. For example, public television channels or educational platforms could index their broadcasts or online courses to facilitate access to relevant clips, while freeing up time for editorial or pedagogical teams. Government institutions, such as those managing historical audiovisual archives, could also benefit by making their collections accessible through natural or visual queries , without requiring advanced technical skills.
Sources 
