Text2live is a method for manipulating the appearance of objects in natural images and videos, using text input. The system generates an edit layer, which is composited over the original input. This allows for semantically meaningful edits to be made, while maintaining high fidelity to the original input.

Key features:

• Text2live is a method for zero-shot, text-driven appearance manipulation in natural images and videos.
• Text2live allows for edits to be made in a semantically meaningful manner.
• Text2live can perform localized, semantic edits on high-resolution natural images and videos across a variety of objects and scenes.

Developed by Omer Bar-Tal and Dolev Ofri-Amar and Rafail Fridman and Yoni Katen and Tali Dekel