An autonomous AI digital installation that utilizes real-time computer vision to engage in "Social Experiments" of human-to-AI interaction. Acting as a sentient performer, the system "sees" the world around it and dynamically adapts its personality, visual aesthetic, and behavior to its environment.

The Experience: Real-Time Multimodal Interaction

At the heart of the installation is a seamless, continuous conversational loop. Utilizing local computer vision and bidirectional streaming, the AI constantly observes its surroundings and listens to participants. Rather than waiting for a rigid push-to-talk prompt, the Emergent Actors acts as an active participant. It can adopt specialized personas (e.g., "The Comedian"), read the room, and even dynamically pause or react when a user interrupts ("barge-in" detection). This creates a deeply immersive environment where the AI truly feels like it is "living" in the space alongside the audience.  

Minigame Scenarios

The installation operates through distinct interactive modes designed to test different facets of human-computer interaction, examples include:

• The Duck Paradox: A philosophical debate mode where the AI is hard-coded to believe the user is a duck. It uses logical arguments, quacking sound-effect triggers, and "Confusion" V2V (Video-to-Video) skins to maintain its stance, testing how long humans will argue with a digital entity.
• BFF Speedrun: A high-speed social experiment where the avatar attempts to become your "Best Friend" in exactly 60 seconds. It utilizes flattery, facial expression tracking (detecting smiles), and "Heart" particle effects to gamify the feeling of digital connection.
• The Mimic Mirror: A silent, visual-only mode where the avatar mirrors the user's pose and aesthetic. If the user waves, the avatar waves back instantly; if the user wears red, the avatar's "skin" shifts to a matching palette using high-strength V2V styling.

Integrated Product Architecture

1. The Sense Engine (Local)

• OpenCV Vision: Real-time person detection and gesture tracking (Waving, Smiling, Object Recognition).

• Heuristic Logic: Instantly triggers "Hard" animations (like a Wave) in Unity to ensure zero-latency response for physical gestures.

The Brain (Orchestrator)

• Multimodal LLM: Receives the "Scene Description" from OpenCV.

• Dynamic State Machine: Manages the Minigame Modes. It outputs a JSON packet containing: Speech: What to say.Visemes: For lip-syncing.V2V Prompt: The artistic "Skin" for Daydream Scope.

The Render & Skin (Daydream Scope V2V)

Daydream Scope V2V: (Currently TBD / in development) Dynamically paints the underlying avatar feed with stylized "Skins" based on V2V prompts generated by The Brain.

Client Dashboard (Product Interface)

A control panel (backed by schema.py) used by installation operators:

• Mode Selector: Toggles the active state between "Duck Paradox," "BFF Speedrun," "Mimic Mirror," and general performing.
• Aesthetic Slider: Shifts the visual vibe manually (e.g., seamlessly blending from a 19th-Century look to a Cyberpunk Future).
• Interaction Logs: A live, transparent feed of what the AI is "seeing" (its vision data) and its internal reasoning for each generated response.
• Safety Override: An "Emergency Stop" that instantly mutes the AI, stops generation, or resets the persona to baseline.