Nodes/ComfyUI MotionDiff/Export Multiple SMPL Subjects toto 3DCGI Software
ComfyUI Node

Export Multiple SMPL Subjects toto 3DCGI Software

Multi-Subject SMPL to 3D Software

By Fannovel16·Created 3 years ago·Updated 2 years ago· 212
Export Multiple SMPL Subjects toto 3DCGI Software
  • smpl_multi_subjects
    foldername_prefix4dhuman_meshes
    formatglb

    When you estimate 3D bodies from video with Human4D_Img2SMPL or faces with SpectreImg2SMPL, you get back an SMPL_MULTIPLE_SUBJECTS object - a stack of every detected person (or face) across every frame, with camera metadata attached. Export Multiple SMPL Subjects to 3DCGI Software is how you get that whole thing out of ComfyUI and into Blender, Unreal, Unity, or any app that eats standard mesh formats. Think of it as the group shot version of the single-subject exporter: instead of one mesh per frame, you get one mesh per subject per frame.

    The output is a folder tree in Comfy's output directory: a top folder for the run, one subfolder per frame, and inside each frame folder a subject_0.glb, subject_1.glb, etc. - one per detected person - plus a camera_intrinsics.json that records the focal length and principal point of the camera the estimation assumed. That JSON is the quiet star of the show: it's what lets you place the meshes back into a 3D scene at the right scale and orientation instead of eyeballing it.

    The inputs that matter

    • smpl_multi_subjects - the multi-subject object from Human4D_Img2SMPL or SpectreImg2SMPL.
    • foldername_prefix - output folder name base (default 4dhuman_meshes).
    • format - mesh format from Trimesh's list; glb is the sensible default.

    It's a terminal node - no outputs, just files on disk.

    Installation

    Part of Fannovel16/ComfyUI-MotionDiff. Install via ComfyUI Manager (search "ComfyUI MotionDiff") or git clone https://github.com/Fannovel16/ComfyUI-MotionDiff into ComfyUI/custom_nodes, install requirements, restart. It's a heavy pack - trimesh does the export work here, and the estimation nodes upstream pull smplx, pyrender, timm and friends.

    Gotchas

    The obvious one is volume: a 10-second clip with three people on screen is 600 mesh files plus a camera JSON per frame. Export only what you need.

    Also remember that if no person is detected in a frame (someone walks off camera), that frame's folder gets skipped - subject_0 in frame 5 isn't necessarily the same person as subject_0 in frame 6. The node doesn't track identities across frames; it just dumps whatever was detected. For clean multi-character work you'll want to sort subjects yourself once the meshes are in your DCC tool.

    And yes, the display name has "toto" in it ("Export Multiple SMPL Subjects toto 3DCGI Software"). That's the pack's typo, not yours - the node works fine.

    CategoryMotionDiff/smpl

    Inputs (3)

    NameTypeDefaultDescription
    smpl_multi_subjectsSMPL_MULTIPLE_SUBJECTS
    foldername_prefixSTRING4dhuman_meshes
    formatCOMBOglb13 options: dae, stl_ascii, xaml, xyz, zae, stl, +7

    Outputs (0)

    No outputs