NGC | Catalog
CatalogModelsStyleGAN3 pretrained models

StyleGAN3 pretrained models

For downloads and more information, please view on a desktop device.
Logo for StyleGAN3 pretrained models


StyleGAN3 pretrained models for FFHQ, AFHQv2 and MetFaces datasets.



Latest Version



August 21, 2023


3.58 GB
This model is backed by NVIDIA’s Plus Plus (++) PromiseClick here to learn more about the quality of the datasets used to train this model
Field Response
Intended Application(s) & Domain(s): Content generation (such as Adobe Photoshop) for generating images of people, animals, etc.
Model Type: Generative Adversarial Network (GAN)
Intended Users: Intended for reproduction of results.
Output: Outputs image
List the steps explaining how this model works: Recognizes different classes of objects
Name the adversely impacted groups this has been tested to deliver comparable outcomes regardless of: Not Applicable
Technical Limitations: Discriminator is not equivariant. Operators in sRGB color space (after tone mapping).
Performance Metrics: Fréchet inception distance (FID)
Potential Known Risks: None Known
Recommended Training:
End User License Agreement: