top of page
Kathrin Federer

DragGAN: Next Level Image Manipulation!

With DragGAN, we don't just edit images, we "drag" them to match our creative vision.


Today we enter the fascinating world of DragGAN, an innovation that revolutionizes image manipulation by integrating AI. With DragGAN, we don't just edit images – we "drag" them to match our vision.


Want to try on clothes on a digital avatar and examine them from all angles? Or do you want to adjust the direction of the pet's gaze in your favorite photo? Or change the perspective on a landscape image? These and similar photo edits, previously reserved for accomplished professionals, are now accessible to amateurs - thanks to a new AI-assisted method that can be implemented with just a few clicks of the mouse.

With DragGAN, anyone can deform an image and precisely control where the pixels go.

Developed by the Max Planck Institute, DragGAN enables interactive control of generative adversarial networks (GANs) by allowing us to "drag" arbitrary points of an image precisely to target positions. Why just read about it when you can see it in action? Check out my demo video below to experience the magic for yourself.



Thanks to AI support, you can adjust the pose, facial expression, direction of gaze or angle on a photo. Doesn't work with your own uploaded photos as of the time of this post, though.


Interested? Try it out for yourself! Just visit DragGAN's official HuggingFace page and follow these simple steps:

  1. Choose a pre-trained model from the dropdown menu.

  2. Choose a seed to create unique images.

  3. Specify two points: Red for the start point and Blue for the end.

  4. Click 'Start' and experience the magic! (Unless it crashes!)

For those who love technical details, you can visit DragGAN's official GitHub page to understand the complex workings and requirements.


Mit DragGAN kann jeder ein Bild verformen und dabei präzise steuern, wohin die Pixel gehen. So können Sie die Pose, Form, den Ausdruck und das Layout verschiedener Kategorien wie Tiere, Autos, Menschen, Landschaften usw. manipulieren.

This revolutionary method is based on Artificial Intelligence, more precisely on "Generative Adversarial Networks" (GANs). GANs are generative models that can synthesize new content such as images. They consist of a generator that creates images and a discriminator that must decide whether the images are real or created by the generator. The system is trained until the discriminator can no longer distinguish the generator's images from real images.


Mit DragGAN kann jeder ein Bild verformen und dabei präzise steuern, wohin die Pixel gehen. So können Sie die Pose, Form, den Ausdruck und das Layout verschiedener Kategorien wie Tiere, Autos, Menschen, Landschaften usw. manipulieren.
©Drag Your GAN: Interactive Point-based Manipulation on the Generative Image Manifold, Xingang Pan, Ayush Tewari, Thomas Leimkühler, Lingjie Liu, Abhimitra Meka, Christian Theobalt (Creative Commons CC BY-NC 4.0)

There are many uses for GANs. For example, besides the obvious image generator application, GANs are good at predicting images. This reduces data overhead when streaming video. They could also upscale low-resolution images and improve image quality.


DragGAN has the potential to revolutionize the way we process images and could have wide-ranging applications in the future, from modifying clothing in photos to producing variations of product presentations to performing various design configurations for planned vehicles with just a few mouse clicks. Although DragGAN works on various object categories such as animals, cars, people, and landscapes, most results to date have been achieved with GAN-generated synthetic images. Applying it to user-entered images is still a challenge that developers are exploring.


The future of image manipulation is pretty crazy and exciting. Tools like DragGAN are pushing the previous boundaries of what is possible.


bottom of page