FLMKR — AI Filmmaking Harness

An autonomous, terminal-native AI filmmaking agent that transforms creative ideas into complete audiovisual productions.

What It Is

FLMKR is an agent harness for filmmaking — not a video generator, not a prompt engine, not a GUI wrapper. It's the environment that allows an autonomous agent to perform complex filmmaking work:

• Plan productions from ideas, screenplays, or treatments
• Create and maintain a Production Bible (characters, locations, style, continuity)
• Generate shots through provider-agnostic adapters (Agnes, Kling, MiniMax, Runway)
• Run QA at chunk/shot/scene levels with continuity checking
• Recover from failures without restarting the entire production
• Render final films with editorial assembly

Quick Start

# Install
pip install -e ".[dev]"

# Create a new film project
flmkr create "A woman discovers her autonomous car is taking her somewhere she never asked to go."

# Plan the production
flmkr plan

# Run production
flmkr run

# Check status
flmkr status

# Resume after interruption
flmkr resume

Architecture

CLI/TUI
  ↓
Producer Orchestrator
  ↓
Agents (Story, Character, Cinematography, Generation, QA, Editorial, Sound)
  ↓
Provider Adapters (Agnes, Kling, MiniMax, Runway)
  ↓
Production Bible (persistent state)
  ↓
Event Log (durable, resumable)

Project Structure

flmkr/
├── __init__.py      # Package entry
├── cli.py           # Click CLI commands
├── config.py        # Pydantic settings
├── models.py        # Core data models (Bible, Shot, Scene, etc.)
├── agents.py        # Agent framework
├── providers.py     # Provider adapters
├── orchestrator.py  # Producer/Orchestrator
├── qa.py            # QA engine
├── editorial.py     # Editorial/timeline engine
├── project.py       # Project state container
├── event_store.py   # SQLite event log
├── events.py        # Event types
├── exceptions.py    # Error types
└── tui.py           # Rich terminal UI

Requirements

• Python 3.11+
• SQLite (bundled with Python)
• Internet connection for provider APIs

License

MIT