Claude Studio Producer

A budget-aware multi-agent video production system that uses competitive pilots to optimize quality and cost.

What Is This?

From a simple idea:

“A day in the life of a writer making a document about how to use a multi-agent system”

To a production-ready platform with:

🎬 7 specialized agents with audio-video orchestration
🎭 Production modes (audio-led for podcasts, video-led for cinematic)
🔌 Pluggable providers (Luma, Runway, DALL-E, ElevenLabs, OpenAI, Google TTS)
🧠 Learning system that improves prompts over time
📚 Knowledge base for document/research-based video production
💰 Budget-aware competitive pilot system with tiered image allocation
🔒 Secure API key management with OS keychain
🎓 Training pipeline for ML-style podcast quality improvement
📽️ Transcript-led video production with scene importance scoring

Example Videos

Coffee Cup - Complete Multi-Provider Pipeline

DALL-E → Luma → ElevenLabs → FFmpeg: Image → Video → Audio → Final Render

Production Pipeline:

DALL-E: Generated coffee cup image from text prompt
Luma: Transformed static image into dynamic video with realistic steam and lighting
ElevenLabs: Created natural voiceover narration with Lily’s voice
FFmpeg: Mixed video and audio with speed-match synchronization

Video Prompt: “Steam rises gently from the coffee cup, morning light shifts slowly across the wooden table, peaceful cozy atmosphere”

Narration: “A perfect morning begins with the gentle aroma of freshly brewed coffee. Steam dances upward in delicate spirals, catching the soft rays of morning light that stream through the window. The wooden table beneath cradles this simple ritual.”

View more examples →

Documentation

📖 Understand the System

Design Evolution - How the system evolved from original vision to current state
Specifications by Timeline - Chronological view of design decisions
Specifications by Theme - Organized by architectural concerns
Developer Journal - Real development notes and lessons learned

🚀 Get Started

GitHub Repository - Source code, installation, quickstart
Examples - Command examples with real outputs
Rendering & Mixing - Video/audio processing features and setup
README - Full project overview

🏗️ Architecture Deep Dives

Core Concepts:

Project Vision - Original “inception” use case and competitive pilot innovation
System Architecture - Data flow and agent orchestration

Agent System:

Producer Agent - Budget planning and pilot strategies
Script Writer - Scene breakdown with provider optimization
Video Generator - Multi-provider video generation
Audio Generator - Multi-provider TTS narration
QA Verifier - Claude Vision-based quality checking
Critic Agent - Evaluation and learning extraction
Editor Agent - EDL creation from best scenes

Provider System:

Provider Architecture - Unified interface, cost estimation, mock mode
Luma Provider - Comprehensive Luma AI implementation
Audio System - TTS production tiers

Advanced Features:

Memory & Learning - Provider-specific continuous improvement
Multi-Tenant Memory - Enterprise namespace hierarchy
Knowledge Base System - Document/research to video pipeline
Knowledge Base Deep Dive - How PDFs become rich knowledge graphs (atoms, topics, figures, quality metrics)
Document Ingestion - PDF atomization and semantic chunking
Podcast Training Pipeline - ML-style iterative quality improvement
Transcript-Led Video Production - Budget-aware visual generation

Quick Commands

# Production (mock mode - no API costs)
claude-studio produce "concept" --budget 5

# Live mode with automatic audio-video mixing
claude-studio produce "concept" --style podcast --budget 10 --live

# Video-led vs audio-led production
claude-studio produce "concept" --mode audio-led --budget 5 --live
claude-studio produce "concept" --mode video-led --budget 5 --live

# Knowledge base workflow
claude-studio kb create "Research" -d "AI papers"
claude-studio kb add "Research" --paper paper.pdf
claude-studio kb produce "Research" -p "Explain transformers" --style educational

# Training pipeline (ML-style podcast quality improvement)
claude-studio training run my-kb-project --reference-audio podcast.mp3
claude-studio training list
claude-studio training show trial_000_20260205

# Video production from training (budget-aware)
claude-studio produce-video -t trial_000 --show-tiers          # See cost breakdown
claude-studio produce-video -t trial_000 --budget low --mock   # Hero images only
claude-studio produce-video -t trial_000 --budget medium --kb my-project --live

# Test providers
claude-studio provider test luma -p "coffee steam rising" -d 5 --live
claude-studio provider test dalle -p "coffee cup on table" --live
claude-studio provider test elevenlabs -t "Your narration text" --voice lily --live

# Manual rendering (advanced - automatic mixing is recommended)
claude-studio render mix video.mp4 --audio narration.mp3 -o final.mp4 --fit speed-match

# Memory/learning management
claude-studio memory list luma
claude-studio memory add luma tip "Use physical descriptions"

# Secure API key management
claude-studio secrets import .env
claude-studio secrets list