The Turing-Complete Canvas is an interactive generative video system that treats video as a stateful software interface rather than a passive clip. Users click objects inside a live AI-generated video — a film reel, a typewriter, a camera — triggering real-time Florence-2 object detection, a two-step LLM orchestration pipeline (via fal's OpenRouter proxy) that authors a dynamic A2UI control-panel surface and rewrites the video-continuation prompt based on the chosen action, and fal's LTX-2.3 extend-video endpoint, which continues the actual footage forward — preserving visual continuity frame-to-frame rather than regenerating from a static image. A Veo 3.1 hero moment delivers a lip-synced character beat at the narrative's climax. Every generation is live, seeded from the actual previous output, with graceful fallback to cached assets if any live call fails. Built solo across the hackathon window.
Team
Placement
Participant
