Researchers from the Max Planck Institute presented an AI model Draggan, which allows you to edit images in near real time. The tool is similar to Photoshop’s warp function, but is much more powerful because the user doesn’t just move pixels, but uses artificial intelligence to recreate the underlying object. In addition, you can rotate images as if they were in 3D.
Image source: arxiv.org / Drag Your GAN
DragGAN can not only resize an object in an image or turn a smile into a frown with a simple mouse click, but also rotate it as if it were a 3D model. For example, you can change the direction of a person’s face. The following videos are from the research team’s website. The full text of the research paper can be read below arXiv.
As the development team points out, what’s really interesting about this work isn’t the image manipulation itself, but the user interface. We have long been able to use AI tools like GANs to create realistic images, but most methods lack flexibility and precision. You can tell the AI image generator:Take a picture of a lion watching the savannah‘ and get it, but it may turn out that the lion is not exactly in the position you need.
DragGAN offers a solution to this problem. The interface is exactly the same as traditional photo editors, but instead of just smearing existing pixels, the model recreates the object. The researchers write:Our approach can not only create an occlusion of objects, but also deform them according to the stiffness, for example when bending a horse’s leg.“.
While this is just a demo and doesn’t fully appreciate the technology, it’s another example of how AI-powered image manipulation can be made more accessible.
Add Comment