Research map / Video / Video generation
CoMo: Controllable Motion Generation through Language Guided Pose Code Editing
Yiming Huang, Weilin Wan, Yue Yang, Chris Callison-Burch, Mark Yatskar, Lingjie Liu
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 LLM-generated fine-grained keywords as conditioning (ablation) | CoMo · HumanML3D test | R-Precision Top-1: 0.487 -> 0.502 | Improved |
| add LLM-generated fine-grained keywords as conditioning (ablation) | CoMo · KIT test | FID: 0.399 -> 0.332 | Improved |
| add fine-grained keywords (ablation) | CoMo · KIT test | Diversity: 11.26 -> 10.95 | Hurt |
| use codebook of 392 pose codes instead of 205 (ablation) | CoMo reconstruction · HumanML3D val reconstruction | FID: 0.059 -> 0.034 | Improved |
| enlarge codebook beyond 392 (661, 733 codes) (ablation) | CoMo reconstruction · HumanML3D val reconstruction | FID: 0.034 -> 0.047 (661), 0.037 (733) | Hurt |
| use downsampling rate 4 instead of 16 (ablation) | CoMo reconstruction · HumanML3D val reconstruction | FID: 0.176 -> 0.034 | Improved |
| reduce downsampling rate from 4 to 2 (ablation) | CoMo reconstruction · HumanML3D val reconstruction | FID: 0.034 -> 0.021 | Improved |
| edit pose codes with an LLM instead of regenerating from edited text | CoMo vs T2M-GPT and FineMoGen · Motion editing user study (54 participants) | human preference: over 70% preferred CoMo on average | Improved |
| use semantic pose codes for generation | CoMo · HumanML3D test | MModality: 1.013 vs 2.799 (MDM) | Hurt |
| use pose-code decoding with generation (vs T2M-GPT) | CoMo · HumanML3D test | inference time per sentence (s): 0.35 -> 0.62 | Hurt |
Later papers that compared with this method
2 later accepted papers report CoMo as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- FineXtrol: Controllable Motion Generation via Fine-Grained Text (AAAI 2026 · better on 4 of 4 shared results)
- LaMoGen: Language to Motion Generation Through LLM-Guided Symbolic Inference (CVPR 2026 · better on 3 of 3 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- FineMoGen — FineMoGen: Fine-Grained Spatio-Temporal Motion Generation and Editing (NeurIPS 2023) · better on 0 of 3 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
- HumanDreamer: Generating Controllable Human-Motion Videos via Decoupled Generation (CVPR 2025 · similarity 0.89)
- InterControl: Zero-shot Human Interaction Generation by Controlling Every Joint (NeurIPS 2024 · similarity 0.88)
- MMM: Generative Masked Motion Model (CVPR 2024 · similarity 0.88)
- Scaling Large Motion Models with Million-Level Human Motions (ICML 2025 · similarity 0.87)
- The Language of Motion: Unifying Verbal and Non-verbal Language of 3D Human Motion (CVPR 2025 · similarity 0.87)
- MotionGPT: Finetuned LLMs Are General-Purpose Motion Generators (AAAI 2024 · similarity 0.87)
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.
