Research map / Video / Video generation
EchoMotion: Unified Human Video and Motion Generation via Dual-Modality Diffusion Transformer
Yuxiao Yang, Hualian Sheng, Sijia Cai, Jing Lin, Jiahao Wang, Bing Deng, Junzhe Lu, Haoqian Wang, Jieping Ye
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add joint video-motion modeling (EchoMotion) vs Wan-5B | Wan-5B · Human T2V benchmark | Human Anatomy: 83.0 -> 85.1 | Improved |
| Add joint video-motion modeling | Wan-5B · Human T2V benchmark | Posture Plausibility (human): 68.9 -> 81.6 | Improved |
| Add joint video-motion modeling | Wan-1.3B · Human T2V benchmark | Human Anatomy: 78.1 -> 79.6 | Improved |
| Train jointly with motion instead of video-only fine-tuning on HuMoVe (ablation) | Wan-5B · Human T2V benchmark | Human Anatomy: 83.1 -> 85.1 | Improved |
| Video-only fine-tuning on HuMoVe (ablation) | Wan-5B · Human T2V benchmark | Human Anatomy: 83.0 -> 83.1 | No clear effect |
| Video-only fine-tuning on HuMoVe (ablation) | Wan-1.3B · Human T2V benchmark | Human Anatomy: 78.1 -> 77.4 | No clear effect |
| Add joint motion modeling | Wan-5B · Human T2V benchmark | Aesthetic Quality: 58.3 -> 58.3 | No clear effect |
| Add MVS-RoPE (synchronized RoPE) (ablation) | EchoMotion · attention-map analysis | temporal alignment | Improved |
| EchoMotion vs closed-source models on complex motion | Veo 3.1 / Kling 2.5 Turbo · Commercial comparison | Video Quality (human): 90.2 / 90.4 vs 80.4 | Hurt |
| Use generative EchoMotion for video-to-motion instead of reconstruction model | HMR2.0 · 3DPW | PA-MPJPE: 58.4 -> 59.8 | Hurt |
| EchoMotion vs earlier generative ChatPose | ChatPose · 3DPW | PA-MPJPE: 81.9 -> 59.8 | Improved |
| EchoMotion vs MotionGPT on motion generation | MotionGPT · Motion generation (50 prompts) | Prompt Following (human): 72.3 -> 82.3 | Improved |
Later papers that compared with this method
1 later accepted paper reports EchoMotion 5B as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- PhyMotion: Structured 3D Motion Reward for Physics-Grounded Human Video Generation (NeurIPS 2026 · better on 2 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Human Motion Diffusion Model (ICLR 2023 · similarity 0.90)
- Large Motion Model for Unified Multi-Modal Motion Generation (ECCV 2024 · similarity 0.90)
- Move-in-2D: 2D-Conditioned Human Motion Generation (CVPR 2025 · similarity 0.89)
- Seamless Human Motion Composition with Blended Positional Encodings (CVPR 2024 · similarity 0.89)
- MimicMotion: High-Quality Human Motion Video Generation with Confidence-aware Pose Guidance (ICML 2025 · similarity 0.88)
- HumanVid: Demystifying Training Data for Camera-controllable Human Image Animation (NeurIPS 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
