patch-gg/anymove

Motion capture for any creature from one AI video: two cameras in one frame, a DINOv2 joint reader, the rig fitted in both views, a GLB out.

What it solves

AnyMove solves the difficulty of animating complex, rigged 3D creatures by turning a single AI-generated video into a fully articulated 3D animation. It eliminates the need for manual frame-by-frame editing by using a two-camera video approach to track joint movement in 3D space.

How it works

The process uses a two-camera setup in a single video frame—one view from the front and one from the side. A custom-trained "joint reader" identifies the creature's joints in both views. Because each joint sits at the same height in both camera angles, the system calculates the exact 3D position of every joint. This data is then used to bend the creature's digital skeleton into the correct pose for every frame, which is finally exported as a GLB file.

Who it’s for

This tool is for 3D artists, developers, and creators who want to animate custom creatures—such as dragons, spiders, or demons—without traditional motion capture hardware or extensive manual keyframing. It is designed for users comfortable with Python, Blender, and using cloud-based AI generation services.

Highlights

  • Converts one AI-generated video clip into a 3D animation in under a minute.
  • Supports complex creatures with wings, tails, and multiple legs.
  • Uses a DINOv2-based vision model to track joints with high precision.
  • Provides a modular pipeline that includes mesh preparation, skeleton rigging, and automated motion capture.
  • Includes "Agent Skills" that allow AI coding assistants to guide users through the setup and troubleshooting process.

Related

  • Project
    spite/FaceMeshFaceGeometry

    A three.js BufferGeometry helper that converts MediaPipe face landmarks into a 3D mesh with head pose, expression scores, and iris tracking.

  • Project
    fayazara/robot-heads

    A React component library that draws animated 3‑D‑style robot heads on a canvas, changing facial expressions and motion to reflect an AI agent’s state.

  • Project
    SpatiaOS/Procedura

    Procedura turns text prompts or images into editable, parametric 3D models by generating OpenSCAD code through an agentic LLM pipeline.

  • Project
    heshuting555/Awesome-3DGS-Applications

    A survey repository on 3D Gaussian Splatting downstream applications, systematically cataloguing datasets and methods for three key directions: segmentation, editing, and generation. Accepted at TPAMI, it is the first dedicated survey covering these 3DGS uses beyond classical view synthesis.