Modulai built an object-detection-based system for detecting highlights and memorable events in online gaming streams, enabling automatic editing of engaging highlight reels. The solution unlocks a capability previously reserved for top-tier streamers with production teams, making it accessible to millions of streamers globally.
Outcome
10
Top titles supported
3
Days to onboard a new title
20
FPS inference
Challenge
A content creator producing a highlight reel from a multi-hour stream typically spends hours finding the best moments and editing them down. For professional top-tier streamers with a team behind them, this is manageable, though expensive and tedious. An automatic solution can unlock this capability to millions of streamers globally and deliver platform-ready content the moment a stream ends.
Solution
A sophisticated pipeline for synthetic data generation and model training was developed in Python, with evaluation, detection, and editing components built in Python and C++. The system was used to onboard some of the most played game titles in the world. It tracks events, evaluates their relevance, builds confidence, and determines intervals to extract and join together into a final highlight video.
Tools
The system uses a synthetic data generation pipeline and object detection models developed in Python and C++. Training, evaluation, and editing are handled through a unified pipeline that can be applied across different game titles with minimal adaptation.
Value created
The solution was used to onboard 10 of the most streamed games in the world from various genres and is capable of accurately detecting and editing videos based on the most important and noteworthy moments in each game.
Learn more
