This is a simple drag-and-drop web interface for generating images with GANs (generative adversarial networks). It supports:
- Generating images from random latent vectors
- Generating variants of an image by moving the latent vector in a random direction
- Interpolating between two images by averaging their latent vectors
- Completing "image analogies" like A:B::C:D (the latent vector of D is calculated as
C+B-A)
A live demo is available here; the model is from ThisJustin-code/pretrained-gan-landscapes-256. You may have to wait a few moments for the container to start up.
The backend is designed to be compatible with NVlabs/stylegan2-ada-pytorch. It should be adaptable to similar architectures by modifying server.py.
You'll need Node.js to build the frontend.
git clone https://github.com/ryanbloom/gan-explorer.git
cd gan-explorer
# Unpickling models relies on having source code available
git clone https://github.com/NVlabs/stylegan2-ada-pytorch
# Download a pretrained model from somewhere
wget $SOME_MODEL_URL -P models
# Bundle JavaScript and run the application
npm install
npx parcel build src/index.html
python server.py