A groundbreaking solution has emerged in a world where complex Photoshop techniques often deter aspiring image editors. DragGAN – a revolutionary tool powered by generative AI – enables users to make significant edits to images through simple point-and-drag controls effortlessly.
Researchers from Google, the Max Planck Institute of Informatics, and MIT CSAIL have outlined the capabilities of DragGAN in a recent paper.
Unlike other generative AI images tools like Dall-E and Midjourney, DragGAN allows users to drop a point on an image, instantly altering its structure and pixels. This unique feature empowers users to achieve precise desired poses and layouts, setting DragGAN apart from its counterparts.
The tool generates realistic results by modifying the angle of an automobile photo, extending the height of a mountain, and turning a closed-mouth lion into one with an open mouth, among other things.
However, DragGAN’s simplicity and intuitiveness are also its greatest advantages. Users don’t need a deep understanding of technology to quickly understand how it works. The interface is focused on giving an image a beginning point and an ending point.
For instance, to create a smile on a person’s face, users can add two points at the corners of the mouth and two additional points slightly further away. With the click of a button, DragGAN seamlessly extends the mouth, while generative AI fills in any gaps to maintain realism. The research paper highlights DragGAN’s ability to hallucinate occluded content, such as teeth inside a lion’s mouth, and deform objects, like the bending of a horse’s leg, based on their rigidity.
Additionally, DragGAN offers a masking feature that allows users to selectively highlight specific parts of an image for alteration while leaving the rest untouched. Apart from its remarkable editing capabilities, DragGAN stands out by enabling users to change the angle from which a photo appears to be taken. While apps like Snapseed offer perspective adjustment, DragGAN takes it a step further by generating pixels from thin air, skillfully filling in gaps that would otherwise require extensive Photoshop work.
DragGAN addresses a significant limitation of traditional image generation tools – their randomized nature. By combining DragGAN with image generation tools, users can achieve outputs that closely match their envisioned images. Although DragGAN is currently available as a demo, its potential applications upon public release are bound to be fascinating.