When we generate images with artificial intelligence, the truth is that we have much less control than we imagine. While we can do quite deep prompt engineering, there comes a moment when we basically have to roll the dice. However, a project called DragGAN presents a new generation of dynamic editing based on point manipulation, allowing the user to alter the appearance, position, and angle of people and objects in an image with relative ease.
Anyone who has experimented with Midjourney, Stable Diffusion, or other image generation models knows that the same prompt can produce very different results. The most effective way to repeat them is by preserving the seed of the image, but if what we're looking for is a slight angle change, a smile, or the correction of a specific error, the prompt is not the best resource to do it. There are too many variables at play, and the model is likely to simply not cooperate.
That brings us to image editing, and a new project called DragGAN promises to be truly revolutionary. The web is already describing it as a "Photoshop killer", or a much more advanced and powerful version of its Warp function. In essence, the user can manipulate the overall condition of an image through a series of points, changing profiles, directions, and styles... but the best way to explain it is with a video.
DragGAN: The Future of Image Editing?
With the assistance of DragGAN, any user can deform an image without sacrificing control over the direction of pixels, effectively modifying aspects such as pose, expression, or the general arrangement of animals, cars, humans, landscapes, and more. The demonstration is sped up, and it makes us suspect that DragGAN requires quite a bit of processing power, but that doesn't stop it from being impressive. Quite the opposite: The user's intervention is reduced to defining manipulation points (red) and target points (blue). Alternatively, there is also the possibility of drawing a mask to limit the flexible region and ignore/isolate the rest of the image.
I imagine it will be necessary to wait a while to see the first precompiled tools, but with such potential, any delay will be worth it. DragGAN goes far beyond classic image editing, changing its essence completely. If you want to see more demos, the link to the official page appears below.
Official announcement and PDF: Click here
