MiniStudio: The Cinematic AI Engine

Programmable, Stateful, and Model-Agnostic Orchestration for High-Fidelity Video Production.

MiniStudio transforms the chaotic world of generative AI into a structured filmmaking pipeline. It solves the "Consistency Problem" by treating video like code—enforcing character identity, environment stability, and temporal continuity through a state-machine driven architecture.

---

See it in Action

The "Why" behind this project and the high-fidelity results (Ghibli 2.0, The Last Algorithm) are documented in detail on my personal site:

Read the Full Article: Programmable Cinematography

---

The Architecture: How it Works

MiniStudio uses a three-layer stack to ensure your characters don't "drift" between shots.

1. Identity Grounding 2.0: We use "Master Reference" portraits (Visual Anchors) that are injected into every injection step, ensuring Emma looks like Emma in Shot 1 and Shot 60.
2. The Invisible Weave: A state-machine that "remembers" the environment geometry. If you move the camera 45 degrees, the engine knows what should be there.
3. Sequential Memory: Each generation is grounded by the final frames of the previous shot, creating a perfect temporal link.

---

Quick Start

1. Installation

pip install -e .

2. Configure Credentials

MiniStudio supports Vertex AI (Veo 3.1) and Google TTS.

Using Doppler (Recommended)

Doppler is a multi-platform secret manager. If you use it, you can run:

doppler run -- python examples/contextbytes_brand_story.py

Using .env or Environment Variables

If you don't use Doppler, simply create a .env file or export your variables directly:

export GOOGLE_API_KEY="your-key-here"
# or
export GOOGLE_APPLICATION_CREDENTIALS="path/to/key.json"

3. Your First Shot

from ministudio import VideoOrchestrator, VertexAIProvider

# Initialize the Director
orchestrator = VideoOrchestrator(VertexAIProvider())

# Define a Shot
shot = ShotConfig(
    action="A lone researcher discovers a glowing orb.",
    characters={"Emma": EMMA_STRICT_ID},
    duration_seconds=8
)

# Produce
await orchestrator.generate_shot(shot)

---

Challenges & Roadmap (AI Filmmaking 2.0)

We are currently pushing the boundaries of what is possible. Current research areas included in our Production Journal:

• Audio-Sync Lag: Refining the waveform orchestrator to eliminate the 0.5s voice/video drift.
• Environment Shimmer: Implementing 2-pass background locking.
• Character Masks: Forcing the AI to paint "over" a locked environment plate.

---

Contributing & Community

MiniStudio is built by the community for the community. See ROADMAP.md for our upcoming features.

Made with love for the future of cinema.