AI Chat Simulation for Prerecorded Videos

Author

Nathan DeVore

Published

June 27, 2025

Back to Portfolio: https://devoreni.github.io/portfolio/

Background

Problem Statement

Live streamers often have a live chat on the screen to keep viewers entertained and chatters engaged; however, managing tens, hundreds, or even thousands of chatters and messages to ensure proper behavior is challenging. If a single bad mannered message makes it through and the live stream is published, it can seriously harm the reputation of the presentor.

Solution

The solution is to prerecord a video, then add fake chatters in post who react to the content as if they were watching live. Generative AI can streamline this process and a custom rendering pipeline can perform the necessary video editing and chat overlay.

Project Requirements

The following requirements are met by the AI Chat project:

  • Context is automatically extracted from the Audio and Video
  • Additional context can be given to enhance responses
  • Context can be modified or changed at any point in the recording
  • Chatters have unique personalities and speech patters
  • Chatters can be pruned from the application
  • Chat messages can be altered in any way

Implementation

Chatter Personalities

Database Structure and Normal Forms

Username Generation

Personalization

Context Gathering

Video Description

Audio Transcription

Previous Chat Messages

User Modifications

Importance of Manual Adjustments

Transcription Editing

Chat message Editing

Removing Chatters

Rendering Pipeline

Frame Extraction

Chat Window Overlay

Built in Log