SAM 3D Body: Full-Body 3D Human Mesh Recovery
SAM 3D Body is Meta's promptable full-body 3D human mesh recovery model with native ComfyUI support for video pose tracking, facial expression driving, and GLB/BVH export.
SAM 3D Body (3DB) is Meta Superintelligence Labs' promptable model for single-image full-body 3D human mesh recovery. It estimates body, feet, and hand pose based on the Momentum Human Rig (MHR) parametric representation, and supports auxiliary prompts such as 2D keypoints and masks. The model is integrated directly into ComfyUI core with SAM3DBody nodes for video pose tracking, facial expression driving, and GLB/BVH animation export.
Usage in ComfyUI
Place the model files in ComfyUI/models/detection/:
sam_3d_body_dinov3_bf16.safetensors
sam_3d_body_dinov3_int8_convrot.safetensorsNo custom node pack is required: the SAM3DBody_Loader, SAM3DBody_Predict, SAM3DBody_Render, SAM3DBody_Smooth, and SAM3DBody_FaceExpression nodes ship with ComfyUI core.
| Developer | Meta Superintelligence Labs |
| Release Date | 2025-11-19 |
| Architecture | DINOv3 encoder + Momentum Human Rig (MHR) decoder |
| License | SAM-License |
| Type | 3D Human Mesh Recovery (HMR) |
| Capabilities | Single-image/video full-body mesh recovery, promptable with keypoints/masks, GLB/BVH export |
Links
- Hugging Face (Comfy-Org): Comfy-Org/sam-3d-body
- Upstream (Meta): facebook/sam-3d-body-dinov3
- GitHub: facebookresearch/sam-3d-body
Guides and workflows related to this model series.
Comments
Sign in with GitHub to join the discussion.