3D Generative AI

A modular Godot framework that routes intent through swappable action nodes, so one controller can drive humans, NPCs, vehicles, and anything else with a consistent behavior API.

MapTrace is a synthetic data machine for spatial reasoning. It teaches multimodal models to respect walls, corridors, and connectivity instead of drawing straight lines through whatever looks shortest.

Inside a C# WPF app that turns Marathon asset extraction into an artist-facing workflow, with one shared backend for both exploration and batch export.

Google Research combines geometry, depth, and a modified Stable Video Diffusion model to turn monocular video into a right-eye view that holds together.

Why writing interpolation algorithms from scratch remains the best way to understand the black boxes of modern computer vision.

A specialized bridge from game rip metadata to production-ready scenes, with shader templates, Geometry Nodes instancing, and engine-specific workarounds baked in.

A custom ComfyUI suite that sidesteps VRAM limits, stitches long videos in chunks, and recovers from bad latents without collapsing the workflow.

A Three.js diorama where stone, storm, sound, and divine intervention are all generated at runtime inside one `index.html`.

StarKnightt turns a 3D image tunnel into a shadcn-style component, using shader math and scroll-driven motion to make a gallery feel alive without a physics engine.

A Cloudflare Worker, a schema shim, and a cost-aware UI turn incompatible image-editing models into a controlled side-by-side test.

A single-file WebGL project uses noise, raymarching-style tricks, and a clean light-mode palette to turn fluid rendering into a portable design artifact.

By intentionally breaking React's cardinal rules of immutability, this browser-based voxel engine bypasses garbage collection spikes to render massive 3D worlds.

A Three.js climbing lab that turns text files into routes, solves reach with Jacobian inverse kinematics, and points toward MuJoCo-level physics when the visualization is not enough.

It keeps gear dimensions in lockstep, draws true involute teeth, and repairs the mesh into something a printer can trust.

A JavaScript and TypeScript engine that measures text once, then reuses cached widths to wrap and size it without leaning on the DOM every time.

A single shader stack turns masks, slots, and lookup tables into dyeable armor, glowing reticles, and layered material behavior that feels far bigger than its codebase.

A tiny Cog package hides a headless OpenGL pipeline, an embedded logo, and a clean way to ship animated brand visuals as one deployable machine.

A bespoke Three.js digital twin that turns one corporate floor into a unit-testable source of truth, with synchronized 2D and 3D views, compliance checks, and a single-file build.

A Three.js system that simulates swells once, reuses the same wave field for rendering and buoyancy, and falls back cleanly when WebGPU is absent.

A bare README turns into a map of the hidden machinery behind diffusion models, and why so many simple AI projects are anything but.

A zero-dependency gradient studio that uses state syncing, blend modes, and one very smart aurora trick to make simple math look cinematic.

A raw WebGL simulation, a glassmorphism control panel, and a reactive theme system packed into a single HTML file.

A photo becomes a five-color palette, the palette becomes a pentatonic score, and the score becomes a new image. The interesting part is not the hallucination, but the constraints that keep the loop readable.

Moving massive architectural models from proprietary desktop vaults to the open web with C# and Three.js.