{"paper":{"title":"Human Motion Diffusion Model","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"A diffusion model for human motion generates natural sequences from text or actions by predicting the clean sample at each step instead of noise.","cross_cats":["cs.GR"],"primary_cat":"cs.CV","authors_text":"Amit H. Bermano, Brian Gordon, Daniel Cohen-Or, Guy Tevet, Sigal Raab, Yonatan Shafir","submitted_at":"2022-09-29T16:27:53Z","abstract_excerpt":"Natural and expressive human motion generation is the holy grail of computer animation. It is a challenging task, due to the diversity of possible motion, human perceptual sensitivity to it, and the difficulty of accurately describing it. Therefore, current generative solutions are either low-quality or limited in expressiveness. Diffusion models, which have already shown remarkable generative capabilities in other domains, are promising candidates for human motion due to their many-to-many nature, but they tend to be resource hungry and hard to control. In this paper, we introduce Motion Diff"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"MDM is a generic approach, enabling different modes of conditioning, and different generation tasks. We show that our model is trained with lightweight resources and yet achieves state-of-the-art results on leading benchmarks for text-to-motion and action-to-motion.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That predicting the clean sample (instead of noise) at each diffusion step, combined with geometric losses, will reliably produce higher-quality and more controllable motions than standard noise-prediction diffusion on the chosen motion datasets and benchmarks.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MDM is a classifier-free diffusion model that generates expressive human motions by predicting clean samples rather than noise, supporting text and action conditioning and outperforming prior methods on standard benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A diffusion model for human motion generates natural sequences from text or actions by predicting the clean sample at each step instead of noise.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"0567be8895ce36e12786533e9d083fe0ccf9df00b682d50a4f6dd0455ceca461"},"source":{"id":"2209.14916","kind":"arxiv","version":2},"verdict":{"id":"d79af2a1-7e01-480e-aade-2c3bf331d15f","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T03:18:23.208904Z","strongest_claim":"MDM is a generic approach, enabling different modes of conditioning, and different generation tasks. We show that our model is trained with lightweight resources and yet achieves state-of-the-art results on leading benchmarks for text-to-motion and action-to-motion.","one_line_summary":"MDM is a classifier-free diffusion model that generates expressive human motions by predicting clean samples rather than noise, supporting text and action conditioning and outperforming prior methods on standard benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That predicting the clean sample (instead of noise) at each diffusion step, combined with geometric losses, will reliably produce higher-quality and more controllable motions than standard noise-prediction diffusion on the chosen motion datasets and benchmarks.","pith_extraction_headline":"A diffusion model for human motion generates natural sequences from text or actions by predicting the clean sample at each step instead of noise."},"references":{"count":20,"sample":[{"doi":"","year":2021,"title":"Accessed: 2021-12-25","work_id":"d78252bc-a5af-4850-872f-ce390aa95f1d","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"A spatio-temporal transformer for 3d human motion prediction","work_id":"7c6af718-79c5-4bbb-bf00-d42954a24f4f","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Text2gestures: A transformer-based network for generating emotive body ges- tures for virtual agents","work_id":"fd721ccb-ee42-4e1d-815f-0ca9cbc7501c","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Implicit neural representations for variable length human motion generation","work_id":"b08e0813-fe00-4320-ad67-024ad0273606","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation","work_id":"af085a26-bb72-4b1a-8e38-107776555081","ref_index":5,"cited_arxiv_id":"1406.1078","is_internal_anchor":true}],"resolved_work":20,"snapshot_sha256":"cf93bbf24451a844a0c6eeca83d2f83f4be8491b3e3a9aa16e5bbad6f2aaad2e","internal_anchors":9},"formal_canon":{"evidence_count":2,"snapshot_sha256":"3bdcdb8100ff192648662a8a312853cac0a789b0b2da1b3bfde1f58f9be03eb9"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}