Adobe has been using AI algorithms in Photoshop for years: being able to remove objects and people or manipulate them within an image has always been possible, but a new AI tool enables exciting new actions.

Investigation. The project is not such an application, it is an academic research by MIT and the University of Pennsylvania in collaboration with Google. The study reveals how this new artificial intelligence tool called DragGAN – the concept of the generative antagonistic network GAN is almost an old acquaintance in the field – helps users manipulate images in ways never seen before.

super vitamin warp. The Distort tool in Photoshop allows for something similar, but with DragGAN this deformation goes much further because that way it doesn’t distort, but rather manipulates images to allow it, as seen in videos. Like shared by @_akhaliqYou can turn the head of a yawning lion or make everyone smile naturally. The researchers also posted several videos on the project’s official website, but the success of the project overloaded the page and many of them crashed.

mind-blowing results. The unique thing about this tool is that it doesn’t actually deform, but instead realistically manipulates these objects. Applying these transformations gives absolutely consistent results and means that we can get the desired result from any original image, for example in terms of pose or expression.

AI creates the image for you, AI manipulates it for you. The tool could be a valuable complement to prolific AI models like DALL-E 2, Stable Diffusion or Midjourney, especially as these create an increasingly spectacular first image, as this tool demonstrates that it is possible to fine-tune this result to achieve exactly the desired result we were looking for.

simple interface. The tool itself offers in its interface a series of controls that also indicate how simple the process is: it is enough to set a few starting points, which is the direction in which the “deformation” will occur, and from there the program will take care of it. create these final images.

Source: DragGAN.

removable. In the image above, you can check out the gait of a horse or potential image changes, ranging from changing the posture and orientation of a dog’s and a cat’s head to the posture of a model (which we can also rotate in 3D). or real-time modification of a car that we can turn into a pickup truck. The changes in landscapes are also surprising, allowing for example the illumination of the scene to be adjusted according to the position of the sun.

on Xataka | Adobe offers AI-based tools that detect if a face is ‘photoshopped’ or remove ‘uhm’ from audio