Key points
- MDM applies a transformer-based diffusion model directly to motion sequences, enabling generation of human motion from text descriptions, action classes, or keyframe conditions.
- Accepted as an ICLR 2023 Oral, MDM is the founding text-to-motion diffusion paper and the origin of a generation of motion-generation tools with direct character-animation application.
- MDM established the text-to-motion diffusion category and underpins motion-generation tools entering character animation and virtual-production pipelines.
Summary
Tevet et al. (2022, published ICLR 2023 Oral) introduce the Motion Diffusion Model, a transformer-based diffusion model that operates directly on motion sequences and accepts conditioning from text prompts, action labels, or partial keyframes. The approach produces diverse, naturalistic human motions without requiring a motion prior or complex post-processing. As the founding paper in text-to-motion diffusion, MDM established the architecture underlying a generation of motion-generation tools that are entering character animation and virtual-production pipelines.
Related items
- DeepPhase: Periodic Autoencoders for Learning Motion Phase Manifolds
- A deep learning framework for character motion synthesis and editing (SIGGRAPH 2016 / SIGGRAPH 2026 Test-of-Time award)
- Iterative Motion Editing with Natural Language
Source
Source: ICLR 2023 (Oral) ↗ (Research)
Cite this item
Tevet, G., Raab, S., Gordon, B., Shafir, Y., Cohen-Or, D. & Bermano, A. H. (2023). ‘Human Motion Diffusion Model (MDM)’, ICLR 2023 (Oral). Available at: https://arxiv.org/abs/2209.14916
Your reference manager can also read this page directly: with the Zotero (or Mendeley) browser connector installed, save it straight to your library. Whole-collection exports: RIS, BibTeX, CSL-JSON.