rinongal/StyleGAN-nada

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1195 stars · 143 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Official PyTorch implementation of StyleGAN-NADA, a CLIP-guided method for adapting pre-trained StyleGAN image generators to new domains using only a natural-language text prompt, with no target-domain training images. It trains a second generator so that the CLIP-space direction between paired outputs aligns with the textual direction.

Use cases

• adapt a StyleGAN generator to a new art style from a text prompt
• convert faces to sketches or paintings without training data
• train an image generator blindly using only CLIP text guidance
• generate domain-shifted images like dogs to lions or churches to huts
• experiment with text-driven GAN domain adaptation in Colab
• invert real photos into an adapted generator's latent space

When to choose

• you want to shift a pre-trained StyleGAN to a new domain described only in text
• you have no images from the target domain to train on
• you need a research-grade implementation of CLIP-guided domain adaptation
• you want to preserve latent-space properties of the original generator

When to avoid

• you need a production image-generation service rather than research code
• you want to train a GAN from scratch on a large image dataset
• you lack a GPU or cannot run PyTorch training
• you need text-to-image generation without a pre-trained StyleGAN backbone

Facets

library · maturity maintenance

machine-learning image-processing deep-learning machine-learning deep-learning artificial-intelligence image-processing python cross-platform stylegan clip domain-adaptation generative-models text-guided gan research-code siggraph-2022 gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="rinongal/StyleGAN-nada")