---

BlendCap reconstructs a full performance, body, hands, and face, from an ordinary video and retargets it onto your Blender rig. No motion-capture suit, no markers, no special hardware, and no second camera. Record on a phone or use footage you already have.

About this repository

This is the public source repository for the BlendCap add-on, published so the source is available as its license requires (BlendCap is GPL-3.0-or-later and includes the AGPL-3.0 YOLO11 detector, see Licensing).

Most people want the paid build: it adds the one-click dependency installer, bundled model access so the setup just works, updates, and support. Building from this source yourself is possible but advanced and unsupported, there are no install scripts here by design; the requirements and model sources are documented in requirements.txt.

---

Contents

• Features
• See it in action
• Requirements
• Get BlendCap
• Documentation
• How it works
• Roadmap
• Licensing
• Acknowledgements

---

Features

• Single-camera, markerless capture: full body and hands from one ordinary clip.
• Finger tracking: both general hand shape tracking and full finger pose tracking, with dedicated controls to tune finger curl or rebuild the finger motion from the tracked keypoints.
• Facial performance capture: drives ARKit-style shape keys and face bones, captured from the same clip or its own.
• Retarget onto your rig: Rigify, Auto-Rig Pro, CloudRig, Mixamo and custom rigs, in both FK and IK, with included presets and a visual bone-map editor.
• Capture cleanup built in: foot grounding and locking, depth-noise filtering, smoothing, tracking-failure repair, and per-take camera-angle correction.
• Runs fully offline after setup, your footage never leaves your machine.
• GPU-accelerated on NVIDIA, with AMD/Intel paths that install automatically but aren't hardware-validated yet, and a CPU-only fallback.

See it in action

Watch the feature demo: YouTube

Requirements

See the full system requirements in the documentation for the GPU tiers and the CPU-only fallback.

Get BlendCap

The easiest way to use BlendCap is the paid build: install it in Blender and click once. It handles everything to get you up and running quickly and effortlessly:

• One-click dependency installer (no terminal, no manual Python)
• Bundled model access, so the ~11 GB setup just works
• Updates and support
• Validated on Windows and Linux with NVIDIA GPUs; AMD/Intel paths install automatically but aren't hardware-validated yet.

Documentation

Full documentation, installation, a first-capture quick start, every setting, and filming tips, lives in docs/ in this repository, start at the table of contents.

How it works

flowchart LR
    V1[video] --> Y[person detection<br/>YOLO11] --> S[body + hands<br/>SAM 3D Body] --> BVH[BVH]
    V2[video] --> M[face landmarks<br/>MediaPipe] --> AK[ARKit shape keys<br/>/ bones] --> BVH
    BVH --> R[retarget onto your rig<br/>FK + IK]

BlendCap runs Meta's SAM 3D Body (a 127-joint body+hand model with a DINOv3 vision backbone, plus MoGe for camera field-of-view) to reconstruct the body, and MediaPipe for the face. Inference is accelerated by the Fast-SAM-3D-Body library on PyTorch / ONNX Runtime. The result is exported as a BVH and retargeted onto your character.

Where it lives in the source: capture runs in save_mhr_data.py / save_face_data.py; the post-processing passes (foot grounding, depth-noise filtering, hand solving, tracking-failure repair, face refinement) in pipeline/; and the retargeting engine (FK bake, IK conversion, constraint simulation, bone-map matching) in blendcap/retarget/.

Roadmap

Directions actively being worked toward, not dated promises, priorities shift with feedback:

Planned

• More rig presets — ready-made bone maps for other popular rigs.
• Smarter auto-match — better automatic bone matching in the map editor, so custom rigs won't need as much manual input.
• Multi-person capture — track more than one performer in a shot and split them onto separate armatures.
• Resumable captures — pause and continue a long capture, and recover a crashed run instead of starting it over from scratch.

Exploring

• Remote GPU capture — run capture on your own cloud GPU when you don't have a local NVIDIA card ("bring your own pod").
• Audio-assisted lip-sync — use the clip's audio to sharpen mouth shapes during speech.
• Dedicated hand capture — film the hands up close for finer finger detail and combine it with a wider body shot, the way separate face capture already works.

Licensing

BlendCap is licensed GPL-3.0-or-later. © 2026 Arcomade. See LICENSE and NOTICE.txt.

Because BlendCap includes the AGPL-3.0 Ultralytics YOLO11 detector as a core, always-loaded component, the whole add-on is distributed under GPL v3 (which bridges to the AGPL component under section 13), which is why this source is, and must remain, public.

Bundled / redistributed components keep their own licenses; full texts are in licenses/:

The animation you create with BlendCap is yours to use, including commercially. The SAM License permits commercial use royalty-free; you're responsible for having the rights to the footage you capture.

Acknowledgements

BlendCap stands on the work of these projects and the teams behind them:

• SAM 3D Body (Meta)
• DINOv3 (Meta)
• Fast-SAM-3D-Body
• YOLO11 / Ultralytics
• MoGe (Microsoft)
• MediaPipe (Google)
• ICT-FaceKit (USC Institute for Creative Technologies)
• PyTorch · ONNX Runtime
• …and Blender and its community.