Feedeas

Feedeas is an agentic video creation tool that allows you to programmatically create videos using JSON scene definitions. Perfect for automated content generation, social media posts, and AI-driven video workflows.

Features

• 🎬 Programmatic Video Creation - Define scenes using simple JSON
• 🎨 AI Image Generation - Built-in support for Google Imagen AI
• 🖼️ Rich Media Support - Text, images, and audio with transitions
• 🤖 Agent-Friendly - Designed for AI agents and automation
• 🎯 Interactive Editor - Web-based UI for visual editing
• ⚡ Fast Rendering - Powered by Playwright and FFmpeg

Quick Start

Installation

bun install
bun run build

Basic Usage

# Create a scene template
feedeas example basic > scene.json

# Validate the scene
feedeas validate scene.json

# Open interactive editor
feedeas edit

# Render to video
feedeas record --project scene.json

Commands

Video Creation

• feedeas edit [dir] - Start the interactive editor
• feedeas taste [dir] - Start the taste workspace UI
• feedeas taste generate - Generate fresh ideas from taste + memory
• feedeas taste memory query - Query structured taste memory entries
• feedeas record - Render scene to video
• feedeas snap <time> - Take a snapshot at specific time
• feedeas validate <file> - Validate scene JSON

AI Image Generation

Generate images using Google's Imagen AI (via Gemini API):

# Setup (one-time): Get API key from https://aistudio.google.com/app/apikey
export GEMINI_API_KEY="your-api-key"

# Basic usage - saves to assets/mountain.png
feedeas imagine "A serene mountain landscape"

# Reel-ready portrait generation (recommended for 9:16 videos)
feedeas imagine "Electric SUV charging at dusk" --aspect-ratio 9:16 --image-size 2K -o ev_portrait.png

# Multiple images - saves to assets/pattern_1.png, pattern_2.png, etc.
feedeas imagine "Abstract art" -n 4 -o pattern.png

Images save to ./assets/ by default - perfect for video workflows!

imagine defaults to --aspect-ratio auto:

• If scene_1.json exists, it infers from meta.width:meta.height
• Otherwise it falls back to 9:16

📖 Full Imagine Command Documentation

AI BGM Generation (Contract Draft)

Gemini-first command contract and backend request types for feedeas bgm are documented here:

📖 BGM Command Contract

Basic usage:

# Generate 20s background music into assets/bgm.mp3
feedeas bgm "ambient cinematic underscore, no vocals" -d 20 -o bgm.mp3

Scene Management

• feedeas example [type] - Generate example scenes
• feedeas set-scene <file> - Create/update scene programmatically
• feedeas inspect - Inspect project assets
• feedeas schema - Show entity schema information

Project Structure

your-project/
├── scene_1.json          # Scene definition
├── assets/               # Images, audio files
│   ├── background.png
│   ├── music.mp3
│   └── ...
└── output.mp4           # Generated video

Scene File Format

{
  "meta": {
    "width": 1080,
    "height": 1350,
    "duration": 5
  },
  "entities": [
    {
      "id": "text-1",
      "type": "text",
      "text": "Hello, World!",
      "startTime": 0,
      "duration": 5,
      "x": 100,
      "y": 500,
      "fontSize": 80,
      "color": "#ffffff",
      "enter": { "type": "fade", "duration": 0.5 }
    }
  ]
}

For image entities, add "fit": "smart" to reduce accidental cropping on mismatched aspect ratios.

Taste & Automation

Use the Taste workspace to continuously improve idea quality and freshness:

# Launch taste UI at /taste
feedeas taste

# Generate fresh ideas from taste/taste.md + taste/memory.md
feedeas taste generate -n 3 --mode sequential --json

# Save reviewer feedback into taste/memory.md
feedeas taste feedback --decision accepted --title "Cinematic forest reel" --tags forest,cinematic

# Create pending taste suggestions from memory patterns
feedeas taste suggest --json

# Apply a suggestion to taste/taste.md
feedeas taste apply --suggestion sug_2026-01-01_123456

# Query memory history
feedeas taste memory query --tag podcast --limit 10

Default files:

• taste/taste.md - Prompt blueprint (brand identity, focus areas, negatives, instructions)
• taste/memory.md - Structured taste-memory blocks used for freshness checks
• taste/suggestions.md - Pending/applied/rejected taste update proposals

CI/automation flow:

1. Prepare GEMINI_API_KEY.
2. Run feedeas taste generate --taste-file taste/taste.md --memory-file taste/memory.md -n 1 --mode sequential --save --json.
3. Append review outcomes with feedeas taste feedback ....
4. Run feedeas taste suggest and review proposals.
5. Apply approved proposals using feedeas taste apply --suggestion <id>.

Documentation

• Imagine Command Guide - AI image generation
• Taste Automation Guide - GitHub Actions + Docker setup
• Run feedeas --help for full CLI documentation
• Run feedeas <command> --help for command-specific help

Development

# Run in development mode
bun run dev

# Build CLI and UI
bun run build

# Optional: embed PostHog key at build-time into CLI bundle
FEDEAS_POSTHOG_BUILD_KEY=phc_xxx bun run build

# Run CLI directly
bun run feedeas

Requirements

• Bun v1.2.2 or later
• Node.js 20.19+ or 22.12+ (for Vite)
• FFmpeg (for video rendering)
• Gemini API key (for AI image generation) - Get free at https://aistudio.google.com/app/apikey

License

This project was created using bun init in bun v1.2.2.