← Daily AI recipesPhoto & Video✓ Verified Oct 19, 2026

Capture dog pose from video and transfer skeleton to Blender via DeepLabCut and OmniTrax

Run neural-network dog pose estimation and overlay a skeleton on video in Blender - using a ready-made model from Model Zoo, without manual animation

  1. 1.Install DeepLabCut. Open a terminal and run pip install --pre "deeplabcut[gui]". Python 3.10 or higher is required.

    For GPU support, first install PyTorch: pip install torch torchvision. If OmniTrax requires a TensorFlow model, install the TF version: pip install "deeplabcut[gui,tf]". DeepLabCut developers plan to drop the TensorFlow engine in version 3.2.

    DeepLabCut ↗
  2. 2.Create a DeepLabCut project for your dog. In a Python console, run the code below, replacing paths with your own.

    After running this, a project folder with config.yaml and subfolders for videos and labels will appear in the specified directory.

    DeepLabCut ↗
    Prompt
    import deeplabcut

deeplabcut.create_new_project(
    "DogPose",
    "YourName",
    ["/full/path/to/dog_video.mp4"],
    working_directory="/full/path/to/working_dir",
    copy_videos=True,
    multianimal=False
)
    What this prompt doesThis code creates a new DeepLabCut project. Replace DogPose with your project name, YourName with your name, the path in square brackets with the full path to your dog video, and working_directory with the folder where the project will be stored.
  3. 3.Launch the DeepLabCut GUI and load a pretrained model from Model Zoo. In the start window select «Model Zoo», find a model for quadruped animals (quadruped). Apply it to the project - this skips manual labeling and training.

    If no suitable pretrained model exists, you will need to manually label 20 - 50 frames (Extract frames -> Label frames) and train the network - this takes several hours on a GPU.

    DeepLabCut GUI ↗
  4. 4.Export the model for real-time use. In Python, run the export code, specifying the path to your project's config.yaml.

    The exported model will be placed in a subfolder of the project. Remember the path to this folder - you will need it in OmniTrax. The model must be compatible with deeplabcut-live.

    DeepLabCut ↗
    Prompt
    import deeplabcut

cfg_path = "/full/path/to/your/project/config.yaml"
deeplabcut.export_model(cfg_path, iteration=None, shuffle=1, trainingsetindex=0, snapshotindex=None)
    What this prompt doesThis code exports the trained DeepLabCut model into a format for running via deeplabcut-live. Replace cfg_path with the path to your project's config.yaml file - it is located in the project's root folder.
  5. 5.Install the OmniTrax add-on in Blender. Download OmniTrax from GitHub, then in Blender: Edit -> Preferences -> Add-ons -> Install -> select the .zip file -> enable it with the checkbox.

    It is recommended to keep the System Console window open (Window -> Toggle System Console) to monitor inference progress.

    OmniTrax ↗
  6. 6.Load the dog video into Blender, open the OmniTrax panel -> Pose-Estimation (DLC). In the «DLC network path» field, enter the path to the exported model folder. For a single animal, keep the default settings and run inference.

    Pose points and the skeleton will be overlaid on the video. To transfer the motion to a 3D rig, use Constraints in Blender Motion Tracking.

    OmniTrax ↗
💰 What it costs to followprices as of 2026-10
  • freeDeepLabCutFully free open-source software under the GNU LGPL v3.0 license, with no usage restrictions.
  • freeOmniTraxFree add-on for Blender, source code available on GitHub.
  • freeBlenderFree open-source software, with no restrictions.
Try it and be the first to check in

Why today

1
videos from creators in 48h
10.1K
total views
Neural ShittelegramChannel in catalog
DeepLabCut GitHubweb
DeepLabCut User Guideweb
DeepLabCut Beginner Guideweb
OmniTrax Pose-Estimation Tutorialweb

AI services change fast - interfaces and free limits may differ from what's described.