Developer Tools

Amazon's New AI Lets You Search Video by Describing It

Find the exact moment in hours of footage — just by asking.

Deep Dive

Amazon just made it possible to search inside video, images and audio the way you search the web — by typing what you want. The company added a model called Marengo Embed 3.0, made by a firm called TwelveLabs, to Amazon Bedrock Knowledge Bases, its cloud service for businesses. That mouthful means: your files get converted into number patterns that capture meaning, so a computer can match a typed sentence to the right moment in a clip.

Until now, building this kind of search required stitching together several separate tools — one to turn speech into text, one to pull out frames, one to store the results. Amazon says the new setup is fully managed: you drop MP4, MOV, JPEG or PNG files into cloud storage and it handles the rest. In its demo, Amazon uploaded a 10-minute clip of the 2022 World Cup final and searched it conversationally.

Who actually cares? Anyone sitting on hours of footage they can't easily search. Sports teams hunting a specific play, teachers looking for a moment in a recorded lecture, security staff reviewing camera footage, retailers checking product videos. Instead of dragging a timeline back and forth, you describe what you remember. Results come back with exact start and end times, so you can jump straight to the right clip.

The catch: this is a tool for businesses with Amazon cloud accounts, not something you can use on your personal camera roll. It also means uploading footage to Amazon's servers, which raises privacy questions for video of faces, homes or workplaces. And it only finds clips — it doesn't edit or share them. It will also occasionally miss or misread a moment, so a human should still double-check.

Key Points
  • You can now search video by describing it — like "the penalty kick in the second half" — instead of scrolling through a timeline.
  • Amazon handles the technical plumbing automatically, chopping footage into 4-second chunks and indexing speech, images and sound.
  • It's a business tool for Amazon cloud customers, so expect enterprise pricing and privacy reviews before your company uploads footage.

Why It Matters

Turns hours of unsearchable footage into something you can question in plain English — a big time-saver.

📬 Get the top 10 AI stories daily