ComfyUI-Sapiens2-Easy
ComfyUI nodes for Meta Sapiens2: segmentation, normal maps, pointmaps, GLB export, and pose.
Nodes (10)
The one knob you'll actually turn
The Sapiens2 Loader That Brings Its Own Weights
Surface normals for people that don't drift off-plane
Surface Normals for a Person, With a Mask Strap Attached
A photo becomes a textured GLB mesh in one node
One Photo of a Person, Exported as a GLB
308 human keypoints, down-mapped to OpenPose when you need it
Sapiens2's Advanced Pose Node
Sapiens2 Segmentation
Segmentation With the Training Wheels Off
ComfyUI-Sapiens2-Easy
From one image to masks, normals, 3D pointmaps, and pose, without leaving ComfyUI.
Build a full human-analysis graph with segmentation, normal estimation, pointmap/GLB export, and pose outputs.
<table> <tr> <td width="50%"><img src="docs/images/sapiens2-pointmap-render-modes.png" alt="Sapiens2 pointmap render modes" /></td> <td width="50%"><img src="docs/images/sapiens2-pose-targets.png" alt="Sapiens2 pose target examples" /></td> </tr> <tr> <td align="center"><strong>Export pointmaps as points, splats, or textured mesh GLB.</strong></td> <td align="center"><strong>Generate pose outputs for BODY_25, DWPose, COCO_18, 308-keypoint, hands, and face.</strong></td> </tr> </table>ComfyUI-Sapiens2-Easy turns Meta Sapiens2 into a small set of ComfyUI-native nodes. Pick a task, model size, and device, then chain the outputs into masks, normal maps, 3D previews, GLB files, or pose JSON.
Why Use It
- One loader for every task: segmentation, normal, pointmap, and pose all share the same model loading flow.
- Auto download and reuse: official Hugging Face weights are saved under
ComfyUI/models/sapiens2/. - Mask-first human workflows: segment the person, then feed that mask into normal or pointmap nodes.
- Single-image 3D export: export pointmaps as points, splats, or textured
.glbmeshes. - Pose outputs for downstream tools: generate
BODY_25,DWPose,COCO_18,308-keypoint, hand, and face targets. - ComfyUI-safe install path: the installer avoids silently replacing your PyTorch/CUDA runtime stack.
What It Supports
| Task | Output |
| --- | --- |
| Segmentation | Body-part masks, merged mask, selected-area preview, raw labels |
| Normal | Sapiens2 surface normal map |
| Pointmap | Depth-style preview and .glb export as points, splats, or mesh |
| Pose | 308-keypoint Sapiens2 pose plus OpenPose-style image and JSON outputs |
Supported model sizes: 0.4b, 0.8b, 1b, 5b
Start Here
| Link | What It Is | | --- | --- | | Install and setup | Full usage guide with install notes, model behavior, node details, and troubleshooting | | Korean guide | 한국어 상세 가이드 | | ComfyUI Registry | Registry listing for this node pack |
Recommended first graph:
Image Load
-> Sapiens2 Segmentation
-> Sapiens2 Normal
-> Sapiens2 Pointmap
-> ComfyUI 3D Preview
Good first settings: model_size = 1b, device = cuda, quality = high, mesh_smoothing = balanced.
Node Set
| Easy Nodes | Advanced Nodes | | --- | --- | | Sapiens2 Model Loader | Sapiens2 Segmentation Advanced | | Sapiens2 Manual Model Loader | Sapiens2 Normal Advanced | | Sapiens2 Segmentation | Sapiens2 Pointmap Mesh Advanced | | Sapiens2 Normal | Sapiens2 Pose Advanced | | Sapiens2 Pointmap | | | Sapiens2 Pose | |
Scope
This repository is a ComfyUI adapter around Meta Sapiens2. It does not train, modify, or redistribute the official Sapiens2 models. Model weights and upstream materials are governed by Meta's Sapiens2 license.
Credits
- Meta Sapiens2: https://github.com/facebookresearch/sapiens2
- ComfyUI: https://github.com/comfyanonymous/ComfyUI
License
This repository's original adapter code and documentation are licensed under the MIT License. See LICENSE.md.
Meta's Sapiens2 models, weights, upstream source code, algorithms, and documentation are governed by Meta's Sapiens2 License: https://github.com/facebookresearch/sapiens2/blob/main/LICENSE.md