Image-Based CLIP-Guided Essence Transfer

We make the distinction between (i) style transfer, in which a source image is manipulated to match textures and colors of target image, (ii) essence one edits include high-level semantic attributes from target. Crucially, that constitute an may differ image. Our blending operator combines powerful StyleGAN generator encoder CLIP novel way simultaneously additive both latent spaces, resulting mechanism guarantees identity preservation feature transfer without relying on facial recognition network. present two variants our method. The first based optimization, while second fine-tunes existing inversion perform extraction. Through extensive experiments, we demonstrate superiority methods for over domain adaptation, text-based editing. code available at: https://github.com/hila-chefer/TargetCLIP .