Researchers from Google, the Max Planck Institute of Informatics, and MIT CSAIL have recently released a new AI technique. It allows users to manipulate images in seconds with just a click and drag. The new DragGAN is an AI editing tool that leverages a pre-trained GAN (Generative Adversarial Network) to synthesize ideas that precisely follow user input while remaining on the manifold of realistic images.

Learn More: An End-to-End Introduction to Generative Adversarial Networks(GANs) 

The Power of DragGAN

DragGAN is an interactive approach for intuitive point-based image editing far more powerful than Photoshop’s Warp tool. Unlike Photoshop, which merely smushes pixels around, DragGAN uses AI to regenerate the underlying object. With DragGAN, users can rotate images as if they were 3D, change the dimensions of cars, manipulate smiles into frowns, and adjust reflections on lakes. Moreover, they can change the direction someone faces.

Also Read: How to Use Generative AI to Create Beautiful Pictures for Free?

General Framework and Optimisation of Latent Codes

What sets DragGAN apart from other approaches is its general framework which does not rely on domain-specific modeling or auxiliary networks. To achieve this, the researchers used an optimization of latent codes that incrementally moved multiple handle points toward their target locations alongside a point-tracking procedure to trace the trajectory of the handle points faithfully. Both components use the discriminative quality of intermediate feature maps of the GAN to yield pixel-precise image deformations and interactive performance.

Outperforming SOTA in GAN-Based Manipulation

According to the researchers, DragGAN by Google outperforms the state-of-the-art (SOTA) in GAN-based manipulation. Furthermore, it opens new directions for powerful image editing using generative priors. They look to extend point-based editing to 3D generative models in the coming months.

Importance of GAN Models

This new technique shows that GAN models are more impactful than pretty pictures generated from diffusion models, such as those used in tools like DALLE.2, Stable Diffusion, and Midjourney. While there are apparent reasons why diffusion models are gaining popularity for image synthesis, GANs saw the same rage and sparked interest three years after Ian Goodfellow proposed them. GAN uses two neural networks—a generator and a discriminator—to generate new and synthesized data instances.

Precise Control Over Image Manipulation

When editing images of diverse subjects, users can “deform an image with precise control over where pixels go. Thus manipulating the pose, shape, expression, and layout,” explains the researchers.

DragGAN: The Future of Image Editing

The DragGAN research paper is the latest example of how AI changes the image editing field. With its user-friendly interface and powerful capabilities, DragGAN has the potential to revolutionize how we edit images.

Also Read: OpenAI Open-Sourced Its Consistency Models for AI Art Generation

Our Say

Google Researchers have unveiled a new AI editing tool for intuitive image editing. Thus, enabling users to manipulate images in seconds with just a click and drag. DragGAN leverages a pre-trained GAN to synthesize ideas that precisely follow user input while remaining on the manifold of realistic images. This groundbreaking research highlights GAN models’ importance and potential for revolutionizing image editing.