New Method for Editing AI-Generated Images by Dragging Them
StableDrag is similar to DragGAN but promises better manipulation.
Image credit: Yutao Cui et al.
We've learned to generate images using AI, now it would be great to edit them without using elaborate prompts. With StableDrag, a new method for AI-generated picture editing, it is possible without too much trouble.
You might find this framework familiar, and it certainly is: last year, we saw DragGAN, a technique that basically does the same. All you need to do is drag a part of the image to the position you need (from the red dot to the blue one in examples), and it will generate an appropriate picture. So what makes StableDrag different?
Image credit: Yutao Cui et al.
Its creators from Nanjing University and Tencent say that their method works better than the others', providing results closer to the original. They designed a "discriminative point tracking method and a confidence-based latent enhancement strategy for motion supervision," which allow them to precisely locate the handle points and get high-quality pictures across all the manipulation steps.
"Thanks to these unique designs, we instantiate two types of image editing models including StableDrag-GAN and StableDrag-Diff, which attains more stable dragging performance, through extensive qualitative experiments and quantitative assessment on DragBench."
Image credit: Yutao Cui et al.
If you want to see how the researchers build StableDrag in detail, check out the paper here. Also, join our 80 Level Talent platform and our Telegram channel, follow us on Instagram, Twitter, and LinkedIn, where we share breakdowns, the latest news, awesome artworks, and more.
