Cartwheel is an AI research lab and product company. We build foundation models for human and object motion, and the tools that put them to work across media and the physical world.

Start free trial
Cartwheel homepage motion preview.

TRUSTED BY

  • Psyop
  • Dogstar
  • Google
  • MR X
  • Bit Reactor
  • Locksmith Animation
  • Snap Inc.
  • FrameMachine
  • The Third Floor
  • BioWare

Scroll horizontally or use the left and right arrow keys to explore all logos.

Our models

Frontier models for the perception, capture and generation of motion, physics and kinematic spatiotemporal data.

Comic

Our family of foundation models for capturing motion from humans and objects to environments

Comic model motion preview.

Swing

Foundation models for generating & editing human motion, rapidly and at scale, from text or poses

Swing model motion preview.

Products

Capture, create and craft motion with precision and scale.

Apps

Performance Generation

Realistic human performance generation from text, at scale

Performance Generation preview.

Pipeline Tools

Cartographer

Universal motion retargeter to robots, 3D characters and other embodiments

Cartographer preview.

MCP

Have ChatGPT, Claude, or your agent capture motion, generate animation, or create rigged characters

Cartwheel MCP works with AI assistants and coding agents.

Research

Model motion, model the world.

The AI systems that will operate within the physical world — humanoid robots, world models, autonomous agents — need training data that reflects that reality.

At Cartwheel, we build the kinetic world models and datasets that make physical AI training tractable at scale.

The human body is the world's most powerful platform and the fastest route to physical AI.

We've discovered that human motion data scales. Models that generate human kinematic spatiotemporal data improve at a predictable rate with increased training data, compute, model size and the correct training approach.

All rapidly scaling data modalities did so after a similar pattern was discovered: language, code, image, audio, video, self-driving. Human kinematics scale in the same regime as language, with 300x more efficiency.

We've built and trained on the largest human motion corpus assembled to date.

Read the research
Loading research scaling graph…

If you're building embodied AI, we'd like to talk.

Try our product suite

Capture motion, generate in batches, retarget and more

Start free trial

Build for custom needs

Custom datasets, models, and evals