DragGAN

Why we included this project

DragGAN is the code behind the SIGGRAPH 2023 paper that lets you edit a GAN-generated image by dragging points on it. You click a few handle points and target points, and the model moves the image content so the handles land where you want them, with an optional mask to keep the rest of the scene fixed. That makes it a practical tool for anyone who needs precise, pixel-level control over pose, shape, expression, or layout in generated imagery, something text prompts can't deliver. Researchers and developers working on generative models or image editing will find the code a clear reference for feature-based motion supervision and point tracking on the generative manifold. It also works on real photos through GAN inversion, so it's useful beyond purely synthetic content. Note that the code ships under NVIDIA's non-commercial StyleGAN3 license, which matters if you plan to use it in a commercial product.