Production Crew

A personal AI-assisted video production studio, built on Cursor agents. Forked from OpenMontage.

What this is

An agentic workflow for creating videos — from concept to final render. The agent handles research, asset generation, prompt engineering, and can control DaVinci Resolve for editing. You direct.

Stack

Layer	Tools
Agent	Cursor (with MCP servers)
Image generation	Recraft V4, Nano Banana 2, GPT Image 2 (via fal.ai)
Video generation	Kling v3/O3, Seedance 2.0 (via fal.ai)
Video editing (Route A)	DaVinci Resolve Studio (via MCP) — full NLE with color, audio, Fusion
Video editing (Route B)	Remotion + HyperFrames + FFmpeg — code-based, no NLE needed
Audio	ElevenLabs, Suno, Piper TTS, ACE-Step

Setup

git clone https://github.com/shuffleo/production-crew.git
cd production-crew
make setup

API keys

Add to .env (all optional — more keys = more tools):

FAL_KEY=your-key              # Image + video generation (Recraft, Kling, Seedance, Nano Banana, GPT Image 2)
ELEVENLABS_API_KEY=your-key   # Voice, music, sound effects
OPENAI_API_KEY=your-key       # GPT Image 2 direct, TTS

DaVinci Resolve — Route A (optional)

cd ~/Github
git clone https://github.com/samuelgursky/davinci-resolve-mcp.git
cd davinci-resolve-mcp
python install.py --clients cursor

Requires DaVinci Resolve Studio (the free edition does not support external scripting). In Resolve 20+, scripting is enabled by default — no preferences toggle needed.

Project structure

production-crew/
├── .agents/skills/        # 64 agent skills (canonical location)
├── .cursor/rules/         # Cursor rules (davinci-resolve, video-art-studio, openmontage)
├── docs/                  # Studio profile, architecture
├── projects/              # Per-project folders (git-ignored)
│   └── <project-name>/   # Each project has its own assets, prompts, and videos
├── tools/                 # Python tool registry
├── skills/                # Pipeline stage directors
├── pipeline_defs/         # Pipeline YAML manifests
└── AGENT_GUIDE.md         # Master agent operating guide

How it works

1. Describe what you want to make
2. The agent picks a pipeline, audits available tools, proposes options with cost estimates
3. You approve, the agent generates assets (images, video clips, audio)
4. Edit in DaVinci Resolve (via MCP or manually) or compose in Remotion
5. Render the final output

Video editing routes

Two paths to a finished video. Pick one per project, or combine them.

Route A — DaVinci Resolve Full NLE. Best for: assembling AI-generated clips, color grading, audio mixing, manual creative editing. Controlled via MCP from Cursor, or hands-on in the UI. Requires Resolve installed with external scripting enabled.

Route B — Remotion + HyperFrames + FFmpeg Code-based. Best for: programmatic composition, motion graphics, data-driven templates, batch rendering. No NLE needed — everything runs from the terminal.

Task	Route
Assemble existing clips into a timeline, grade, mix audio	A — Resolve
Generate motion graphics, animated titles, captions	B — Remotion/HyperFrames
Combine AI clips + audio into a final music video	A — Resolve (or B if fully scripted)
Batch render templated videos with different data	B — Remotion
Ambiguous	Ask the user

License

GNU AGPLv3

---

Forked from OpenMontage by calesthio.