mlx-generate-ctmr

Native MLX weights for NVIDIA NV-Generate CT, paired CT mask generation, general MR and MR-Brain. These weights were converted to safetensors without retraining.

Directory	Configuration	Contents
`ct/`	CT	Image VAE, UNet and ControlNet, plus the paired-CT label mapping and anatomy-size database
`mask/`	Learned paired CT masks	Mask VAE and UNet
`mr/`	General MR	Image VAE and UNet
`mr-brain/`	MR-Brain	Image VAE and UNet

Use the mlx-generate-ctmr package and pass the appropriate directory to its runner. Learned paired CT uses ct/configs/label_dict_124_to_132.json from NV-Generate-CTMR (Apache-2.0) and ct/datasets/all_anatomy_size_conditions.json from NV-Generate-CT, both unmodified. MR configurations require the conversion.json beside model.safetensors.

Quick start

uv venv --python 3.13
source .venv/bin/activate

uv pip install "mlx-generate-ctmr>=0.5.2.post2"
uv tool install huggingface-hub

hf download MedMLX/mlx-generate-ctmr --revision f9ce3d2b09fd81e24e54e00e1445637805581733 --include 'mr-brain/*' --local-dir models

Replace the example input paths with your own files.

from mlx_generate_ctmr.medmlx import run

result = run(
    configuration='mr-brain',
    modality='mri_t1',
    weights_path='models/mr-brain',
    output_dir='outputs',
    seed=0,
)
print(result["outputs"])

For CT, download ct/* and mask/* and pass weights_path='models/ct', adding mask_weights_path='models/mask' for paired generation.

Use & license

For research and education. Not validated for clinical use. Synthetic output is not ground truth.

General MR is for noncommercial research only under the NVIDIA OneWay Non-Commercial License. The MR-Brain UNet comes from NV-Generate-MR-Brain, and the CT and paired-mask weights come from NV-Generate-CT. MR-Brain also uses the CT VAE.

NVIDIA's CT model card metadata identifies the Open Model License, while its terms-of-use link points to the Community Model License. Until NVIDIA clarifies the CT terms, comply with both the Open Model License and the Community Model License for CT, paired-mask and MR-Brain bundles. The Community agreement restricts some uses of outputs to develop other AI models and conditions production use; consult the agreements. See the root NOTICE, upstream notices, and the ct/NOTICE and mask/NOTICE for source revisions and attribution.

References

Citation

If you use MedMLX, please cite the MedMLX paper:

@article{sandoval2026medmlx,
  title   = {MedMLX: On-Device Medical Imaging AI for Apple Silicon},
  author  = {Sandoval, A. and Sandoval, J. and Saleem, A.},
  journal = {medRxiv},
  year    = {2026},
  doi     = {10.64898/2026.10.28.26366762}
}