# MoCapAnything V2 maps monocular video to joint rotations for any skeleton without inverse kinematics

The authors report rotation error of about 10 degrees, against 17 to 20 for analytical pipelines, and 20 times faster inference, with code and weights under the MIT license.

- Published: 2026-10-29T06:14:22.974Z
- Canonical: https://polylog.news/ai/2026-10-29/mocapanything-v2-maps-monocular-video-to-joint-rotations-for
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Polylog editors](https://polylog.news), [arXiv](https://arxiv.org/pdf/2604.28130), [Hugging Face](https://huggingface.co/kehong/MoCapAnythingV2-weights)

The authors of MoCapAnything have released version 2, an end-to-end motion-capture system for arbitrary skeletons. It takes a monocular video and a one-frame reference pose and outputs joint rotations ready for the Biovision Hierarchy (BVH)…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-10-29/mocapanything-v2-maps-monocular-video-to-joint-rotations-for (subscription information: https://polylog.news/pricing).