AWS Elemental Inference Converts Live Video for Mobile Viewers

AWS Elemental Inference is a fully managed AI service that automatically optimizes live and on-demand video, converting broadcasts into formats for mobile and social platforms in real time. At launch, it supports vertical video creation and clip generation, enabling content tailored for channels like TikTok, Instagram Reels, and YouTube Shorts without separate post-production work.

Previously, converting horizontal broadcast footage into vertical format required a lot of manual editing, making it easy to miss viral moments. This service can be configured independently through the AWS Management Console or attached directly to an AWS Elemental MediaLive channel to add AI capabilities to existing live workflows.

The main usage flow is as follows.

• Standalone console: Use Create feed to create the parent resource, then configure crop or clip generation outputs.
• MediaLive integration: Enable AI features in channel settings and manage output configuration, including Smart Crop.
• Centralized management: Check the ARN, data endpoint, and output status in the dedicated AWS Elemental Inference tab on the MediaLive channel detail screen.

In terms of how it works, agentic AI automatically performs analysis and optimization on live video. Vertical video cropping and clip generation run independently of each other and handle multi-step conversion without separate human-in-the-loop prompts.

Latency is on the order of 6-10 seconds, faster than the minutes-long work required by existing post-processing methods, following a process once, optimize everywhere approach that applies multiple AI features to a single stream simultaneously. The service uses foundation models (FMs) managed by AWS to provide automatic updates and optimization, so it can be operated without a dedicated AI team.

Launch features are as follows.

• Vertical video creation: Automatically crops to a 9:16 ratio, converting footage into a format suited for mobile and social platforms.
• Clip generation with advanced metadata analysis: Automatically extracts highlights from live broadcasts to help with real-time distribution.

It is currently available in 4 AWS Regions: US East (N. Virginia), US West (Oregon), Europe (Ireland), Asia Pacific (Mumbai). Pricing is consumption-based pricing, where you pay only for the features used and the video processed, with no upfront cost or commitment.