Swing is our family of generative human motion models

Swing generates human motion from text and poses. Describe the action, intent, style, timing, looping, and number of performances, and Swing produces them rapidly and at scale. Trained on the largest, highest-quality motion dataset ever assembled, Swing brings real-world knowledge to the motion it creates.

By generating smooth motion in seconds, Swing produces motion datasets at scale, far faster and more cheaply than ever before.

For teams needing more specificity than a general model can offer, we build custom versions of Swing tuned to a single character, a style of movement like a kung-fu fighter or a wounded soldier, or a whole cast that shares a common language of motion.