RealityCompiler

• 

  Parabolic Projectile

• 

  No Ghost Trail

• 

  Inclined Ramp Drop

• 

  Negative Gravity + Bounce

• 

  Orbital Pendulum

• 

  Spawned Spheres + Ghost Trail

• 

  Positive Gravity + Bounce

Inspiration

I kept seeing AI tools restricted to static chat boxes, PDF summarization, or generic text generation. I wanted to build something that bridges the gap between the physical world and digital code—transforming passive video pixels into active, tweakable math that judges could physically interact with in seconds.

What it does

RealityCompiler is an AI-powered video-to-physics simulation engine. When I drop any short video clip—such as a ball rolling down a ramp or an object in motion—the system extracts its visual trajectory and symbolic physics equations, compiling it into a live, interactive 2D simulation where I can adjust gravity, friction, and bounciness in real time.

How I built it

I designed the interface using Next.js, Tailwind CSS, and Lucide icons to achieve a clean, dark-mode precision engineering aesthetic. For the core execution layer, I engineered a self-contained, high-performance 60fps HTML5 Canvas physics loop that handles rigid-body collisions, mouse-dragging, and live parameter adjustments with zero external dependency bloat.

Challenges I ran into

Bridging static video frames with deterministic runtime math presented a unique challenge. Ensuring that custom object spawning, gravity toggles, and restitution formulas stayed smooth and responsive inside a standard React component required fine-tuning the animation frame loop and state management to prevent rendering jitter.

Accomplishments that I'm proud of

I successfully built a working, highly polished interactive proof-of-concept that turns video pixels into tweakable physics mechanics instantly, delivering an intuitive "engineering studio" feel right out of the box.

What I learned

I deepened my technical grasp of combining multimodal vision concepts with deterministic physics engines, optimizing HTML5 canvas render loops, and designing micro-tool interfaces for rapid AI workflows.

What's next for RealityCompiler

I plan to wire up direct multimodal API support (such as Gemini Pro Vision) for the upload pipeline, allowing users to drop any random phone video and instantly generate real-world matched physics simulations on the fly.

Built With

• canvas
• css
• html5
• javascript
• next.js
• node.js
• react
• tailwind
• typescript
• vercel