Skip to content

Installation

Requirements

  • Python 3.9 – 3.12
  • PyTorch ≥ 2.1 (CPU or CUDA; Apple-silicon MPS models also work, since tensors are summarized on the model's own device)
  • Core dependencies, installed automatically: numpy, matplotlib, pillow, networkx

Install from source

git clone https://github.com/MASILab/pymodelvis.git
cd pymodelvis
python -m venv .venv
source .venv/bin/activate            # Windows: .venv\Scripts\activate
pip install -e .

The distribution is named pymodelvis and the import name is neural_flow:

import neural_flow
print(neural_flow.__version__)

It also installs the neural-flow command (see the command-line guide):

neural-flow --version
neural-flow demo cat      # writes neural_flow_demos/demo_cat.png

If your shell can't find neural-flow, the virtual environment isn't active or pip's script folder isn't on PATH. python -m neural_flow … always works.

Optional extras

extra installs needed for
examples torchvision, timm, scikit-image, torchxrayvision the example scripts (ResNet, ViT, chest X-ray, sample photos)
animation imageio, imageio-ffmpeg MP4 output (GIF works without it)
medical monai, nibabel, einops, nilearn, dynamic-network-architectures, scipy MONAI bundles (UNesT), nnU-Net models (TotalSegmentator), NIfTI I/O, the MNI152 template
dev pytest, torchvision, monai, nibabel, dynamic-network-architectures, scipy running the tests (including the 3-D transformer and nnU-Net tests)
all everything above 
pip install -e ".[examples,animation]"
pip install -e ".[all]"

CPU-only PyTorch (smaller download on Linux)

pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu
pip install -e ".[all]"

Verify

pytest -q                 # CPU only, about a minute
neural-flow demo cat      # or: python examples/resnet.py → examples/outputs/resnet50_cinematic.png

The first run downloads ResNet-50 weights (~100 MB) once. The command-line tool caches them in ~/.cache/neural_flow (or $NEURAL_FLOW_HOME); the example scripts use real_models/, which is git-ignored.

Fonts

The cinematic style uses the Inter typeface, which ships inside the package (neural_flow/fonts, SIL Open Font License). Nothing needs to be installed system-wide.