DragGAN

DragGAN is an interactive image editing framework that enables users to precisely manipulate objects within images by “dragging” points to desired target locations. Built on top of generative adversarial networks (GANs), DragGAN operates in the latent space of a pretrained generator, allowing edits that maintain realism, coherent structure, and consistent lighting while significantly altering pose, shape, or expression.

The tool provides point-based controls: users place handle points on source image regions and target points where those regions should move. DragGAN then iteratively optimizes the latent code so that the selected features follow the user-defined motion while preserving object identity and fine details. This supports complex edits such as changing facial expressions, adjusting body posture, reshaping cars or animals, and modifying object geometry without manual masking or 3D modeling.