Condé Nast says an AI-powered search system built with AWS cut the time its editorial teams spent finding video from an average of 250 minutes to about two minutes per task. The system searches across more than 140,000 videos by what is seen, said and heard, returning relevant clips with timestamps.
AWS AI Watch analysis
What happened
Condé Nast and AWS built the system using Amazon Bedrock, Amazon OpenSearch Service and TwelveLabs’ Marengo embedding model. In a May 2026 benchmarking workshop, Condé Nast measured a 99.2 per cent reduction in discovery time and more than a 90 per cent reduction in manual video review. The figures and estimated annual savings of about $800,000 are Condé Nast’s measurements and estimate, presented in AWS’s account of the project.
The tool offers more than a smarter keyword box:
Our top picks
- Search by intent
Editors can describe a clip in natural language and get relevant moments with precise timestamps. - Search across sound, speech and pictures
Results can match what a video shows, what is said in its transcript or what is heard. - Use an image as the query
Editors can upload a reference image to find visually similar material in the archive. - Handle imperfect searches
The search layer is designed to resolve typos and focus on meaning rather than exact wording. - Keep ingestion separate from searching
The architecture processes and indexes video asynchronously, while serving queries through a separate system.
Why it matters
For an editorial team working across Vogue, GQ, Vanity Fair and Wired, finding an old clip is not just a filing problem. Faster discovery could help staff respond to requests and put previously overlooked archive footage back to work. The reported time saving is striking; the estimated savings and business benefits remain Condé Nast’s own assessment, not a general guarantee for other archives.
Our read
This is a compelling example of multimodal AI solving a recognisable workplace nuisance: searching for a moment in a library too large to browse by hand. The useful detail is that the system finds specific video segments across image, audio and transcript signals, rather than merely dressing up a catalogue search. Other organisations should treat the results as a promising case study, then test against their own footage and the queries their staff actually make. Even a very clever search box cannot rescue an archive nobody has indexed properly.
What to watch
- Whether Condé Nast shares results from use beyond its May benchmarking workshop.
- How reliably search works across different video types and real editorial queries.
- Whether the estimated savings are borne out as teams use the system over time.
Discussion spark: Would you trust AI search to surface the right moments from a huge video archive, or should editors still verify every result before using it?
Sources and evidence
- How Condé Nast built multimodal video discovery with Amazon Bedrock (29 September 2026, 15:55 UTC)
not affiliated with or endorsed by Amazon Web Services (AWS)