Tag
The paper presents Puppeteer, a diffusion-based co-speech gesture model that uses causal latent tokens and object geometry to generate temporally coherent, posture-aware, and physically grounded gestures.