Manipulating Images like Magic – DragGAN
The art of image manipulation is undergoing a revolution thanks to AI, which is expanding the boundaries of what can be achieved. With this tool, users can manipulate images by dragging points to close open eyes, alter clothing, or add elements, all while maintaining a realistic appearance that closely resembles the original image.
DragGAN, represents a significant leap forward in image editing technology. It’s not a publicly available tool but rather exists as a research paper developed by researchers from Google, the Max Planck Institute of Informatics, and MIT.
By utilizing a pre-trained generative adversarial network (GAN), DragGAN ensures that edits stay within the realm of realism. This means that even significant alterations to the original images will retain a natural and authentic look, avoiding the pitfalls of exaggerated and unnatural visual effects.
DragGAN’s impressive capabilities are the result of adversarial training, where the generator and discriminator components of the network work together to refine image generation and discrimination. The generator’s objective is to create images so realistic that even the discriminator, trained to distinguish real from fake, cannot tell them apart. Through this iterative process, both components improve their performance, pushing the boundaries of what DragGAN can achieve.
For a user, this can in practice mean opening the mouth of a person’s face in an image, prolonging a short dress to a long dress, or changing the pose of a body – and making it look as realistic as the original image.
DragGAN or similar tools’ potential impact on the established leader in the field, Adobe Photoshop, is significant, as it introduces entirely new possibilities for editing and manipulating images.
