Core Concepts
GANTASTIC is a novel framework that transfers interpretable directions from pre-trained GAN models directly into diffusion-based models to enable disentangled and controllable image editing.
Abstract
The article introduces GANTASTIC, a novel framework that aims to combine the disentangled editing capabilities of Generative Adversarial Networks (GANs) with the generative excellence of large-scale text-to-image diffusion models.
The key highlights are:
GANTASTIC is the first approach to transfer directions from a pre-trained GAN model to a pre-trained text-to-image diffusion model without finetuning.
The framework can transfer a wide range of fine-grained directions spanning various categories, including faces, cats and dogs.
The transferred directions are highly disentangled and can be applied together without interfering with each other.
Experiments show that GANTASTIC achieves disentangled editing results that are competitive with state-of-the-art diffusion-based and GAN-based image editing techniques.
The authors share the source code and discovered directions to enable further research in this area.
Stats
The article does not contain any key metrics or important figures to support the author's key logics.
Quotes
The article does not contain any striking quotes supporting the author's key logics.