Title

Stream Video Inference Demo: AI-Powered Content Analysis

Summary

Taylor, Product Manager for Stream, demonstrates a new feature for on-demand video bindings and event subscriptions. The demo showcases video inference capabilities, including summarization, highlight reel creation, ad break identification, and content moderation. The system uses a workflow involving Stream, Queues, Workers, D1, and external AI services like 12 Labs Pegasus and AWS Nova 2. The demo highlights AI-generated content analysis, including titles, summaries, chapter segmentation, and object detection. Taylor proposes further collaboration with 12 Labs and internal teams to enhance customer experience.

Preview Frame

Generated by sending the URL of a thumbnail/still frame from 50% of the video duration.

AI Generated Content and Deepfake Analysis

Visual Moderation Analysis

This site doesn't have much scandalous content, but for every no_* class, there is an inversely corresponding yes_* class for the same type of content. Each will have a 0 – 1.0 value to indicate the probability that sample contained (or not) that class of content.