ChatPose
A multimodal LLM that chats about 3D human pose — estimation and generation from text or images.
About
ChatPose embeds 3D human poses as discrete signal tokens inside a multimodal large language model, so you can literally chat about bodies: ask what someone in a photo is doing, request a 3D pose from a textual description, or get speculative pose generation from subtle queries. A specialized SMPL projection layer converts the model's language embeddings into SMPL pose parameters and full body meshes.
The CVPR 2024 paper unifies classical pose estimation and generation with two new reasoning tasks — speculative pose generation and reasoning about pose estimation — and reports that ChatPose beats both existing multimodal LLMs and task-specific methods on those benchmarks. Code and data are available for research. It is a research prototype, not a product: there is no hosted demo or API.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Provider-reported (paper, arXiv 2311.18836): outperforms existing multimodal LLMs and task-specific methods on the two newly proposed pose-reasoning benchmarks
- Benchmarks are bespoke to the paper; no standard leaderboard comparison
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (TIER metrics, metrics-v1). Evidence: CVPR 2024 peer review (arXiv 2311.18836, MPI-IS/ETH/Meshcapade team); code+data released for research. Strengths: strong venue, genuinely novel LLM-pose interface. Docked for: (1) bespoke benchmarks with no independent replication; (2) research prototype with no weights/API and unclear license. License: not stated.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 5.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 5.0/10
- Manual review · sanity read + red-flag check
- 6.0/10
- · CVPR 2024 peer review (arXiv 2311.18836), MPI-IS/ETH/Meshcapade team
- · Code + data released for research at yfeng95.github.io/ChatPose
- · Benchmarks bespoke to the paper; no independent replication found