Skip to content

tinycrops/starter-applets

 
 

Repository files navigation

Generated Image

Look at this video for an explaination of the Short Term Memory, Long Term Memory Working Memory query feature https://x.com/tinycrops/status/1910105664589799515

Video Watcher - Gemini Dataset Builder

This application watches a folder for new video recordings from OBS, sends them to Google's Gemini AI for analysis, and builds a dataset of AI-generated labels and descriptions.

Features

  • Automatic Video Detection: Monitors a specified folder for new video recordings
  • AI Analysis: Sends videos to Gemini for detailed analysis
  • Dataset Building: Creates a structured dataset of AI-generated labels
  • Web Interface: View and explore the generated dataset

Prerequisites

  1. Node.js 18.x or higher
  2. Google Gemini API key
  3. OBS Studio configured to save recordings to a specific folder

Setup

  1. Clone this repository
  2. Install dependencies:
    npm install
    
  3. Create a .env file in the project root with the following variables:
    VITE_GEMINI_API_KEY=your_gemini_api_key_here
    VIDEO_WATCH_FOLDER=Q:\\
    VIDEO_DATASET_FOLDER=C:\\Users\\YourUsername\\video-dataset
    
    Note: Update the paths to match your system configuration.

Usage

  1. Start the application:

    npm run dev
    
  2. The server will start watching the specified folder for new video recordings

  3. Record a video in OBS and save it to the watched folder (Q:\ by default)

  4. The application will automatically detect the new video, send it to Gemini for analysis, and add it to the dataset

  5. Open your browser to http://localhost:8001 to view the web interface

How It Works

  1. The application uses chokidar to watch the specified folder for new video files
  2. When a new video is detected, it is uploaded to Google's Gemini AI
  3. A structured prompt asks Gemini to analyze the video and provide detailed information
  4. The response is parsed and saved to the dataset folder as a JSON file
  5. The web interface displays all analyzed videos and their AI-generated metadata

Customization

  • To modify the prompt sent to Gemini, edit the DEFAULT_PROMPT in server/video-processor.mjs
  • To change the watched folder or dataset location, update the environment variables in .env

Folder Structure

  • /server: Backend Node.js server code
  • /src: Frontend React application
  • /server/video-processor.mjs: Core module for video analysis with Gemini

About

A Memory module for users who want to feel like their AI companion understands them

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • JavaScript 91.8%
  • CSS 7.9%
  • HTML 0.3%