How to Use SMPL Format Motion Capture in QuickMagic

Overview: SMPL Output Now Available

We are excited to announce that SMPL format motion capture is now live on QuickMagic. When you capture motion from a video, you can now select SMPL as your output format, alongside our existing export options.

SMPL (Skinned Multi-Person Linear Model) is the most widely used parametric 3D human body model in research and industry. Choosing SMPL output means your captured motion is delivered as standard SMPL parameters — ready for computer vision research, 3D reconstruction pipelines, game development, and any workflow built on the SMPL ecosystem.

Your captured result is delivered as a single .npz file containing per-frame SMPL pose and shape parameters.

How to Capture: Get Your SMPL File

1. Upload your video. On the Motion Generation page, upload the video you want to capture motion from.
2. Select the SMPL format. In the output format options, choose SMPL.

3. Submit the task and wait for processing. Our AI solves body motion from your video and converts it into standard SMPL parameters.
4. Preview the motion. Check the captured result in the viewer.
5. Download the npz file. If the result meets your expectations, download the .npz file to your local machine.

What Is Inside the npz File

The npz file you download contains four NumPy arrays — the standard SMPL parameter set:

How to Use Your SMPL npz File

Step 1: Download the SMPL Model

1. Register at https://smpl.is.tue.mpg.de/register.php and agree to the license agreement (free for academic research).
2. Log in with your account. Access may require email confirmation and is not activated instantly.
3. Go to the download page: https://smpl.is.tue.mpg.de/download.php (or click "Downloads" in the top menu).
4. Download SMPL for Python and unzip it. You will find:
   • SMPL_NEUTRAL.pkl (gender neutral — this one is enough)
   • SMPL_MALE.pkl / SMPL_FEMALE.pkl (optional)

Step 2: Set Up the Python Environment

After installing Anaconda, open a terminal and run:

conda create -n smpl_env python=3.9 -y
conda activate smpl_env
pip install torch
pip install smplx
pip install trimesh chumpy

Step 3: Organize Your Folder

your_work_folder/
├── your_mocap.npz           <- the npz file downloaded from QuickMagic
├── smpl_models/
│   └── smpl/
│       └── SMPL_NEUTRAL.pkl <- the model from Step 1
└── run_smpl.py              <- the script from Step 4

Step 4: Run the Script and Generate the 3D Body

Create a file named run_smpl.py with the following content:

import numpy as np
import torch
import smplx
import trimesh
import os

# 1. Load your npz data
data = np.load("your_mocap.npz")
num_frames = len(data["transl"])
print(f"Total frames: {num_frames}")

# 2. Load the SMPL model
model = smplx.create(
    model_path="smpl_models/",
    model_type="smpl",
    gender="neutral",
    batch_size=num_frames
)

# 3. Feed in the pose parameters to compute the body mesh
output = model(
    global_orient=torch.tensor(data["global_orient"], dtype=torch.float32),
    body_pose=torch.tensor(data["body_pose"], dtype=torch.float32),
    transl=torch.tensor(data["transl"], dtype=torch.float32),
    betas=torch.tensor(data["betas"], dtype=torch.float32),
)

vertices = output.vertices.detach().numpy()  # (frames, 6890, 3)
faces = model.faces

# 4. Export each frame as an obj file
os.makedirs("output_meshes", exist_ok=True)
for i in range(num_frames):
    mesh = trimesh.Trimesh(vertices=vertices[i], faces=faces)
    mesh.export(f"output_meshes/frame_{i:04d}.obj")

print("Done! All frames exported to the output_meshes/ folder.")

Then run:

python run_smpl.py

Once finished, the output_meshes/ folder will contain one obj file per frame.

Step 5: View the 3D Results

• MeshLab (free): double-click any obj file to view a single frame.
• Blender (free): File → Import → Wavefront (.obj); enable sequence import to load all frames and play the animation.
• Windows 3D Viewer: double-click an obj file directly.

FAQ