Skip to content
Book Open access

COME: Advancing Motion Representation and Generative Modeling for High-Quality Text-to-Motion

Nov 2026 · Proceedings of the 34th ACM International Conference on Multimedia · 0 citations · 50 references

Abstract

Text-to-Motion generation aims to synthesize 3D human motion from natural language. While diffusion models naturally capture motion continuity, they often underperform discrete token-based approaches. Emerging applications such as motion editing and multimodal control, however, require fine-grained control, compositionality, and diverse sampling, which are better supported by continuous frameworks. Motivated by this gap, we conduct a systematic analysis of existing continuous diffusion methods and identify two key challenges. Prior continuous approaches fall short in both motion representation and generation, exhibiting limited discriminative quality in latent spaces and suboptimal performance in reconstruction and synthesis. To address these challenges, we propose COME, a continuous diffusion framework that jointly improves motion representation and generation. It integrates a high-quality motion tokenizer, MoCMAE, which produces more discriminative and structured features, with a diffusion generator, ccDIT, that stabilizes training and inference while preserving fine-grained semantic alignment. Extensive experiments show that COME achieves state-of-the-art generation quality and efficiency, outperforming both prior continuous and discrete approaches.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.