High-Dimensional Feature Fusion for Font Generation with Denoising Diffusion

When writing a vast number of characters, and lacking consist font, generating define font is a challenge problem. In order solve to this problem, this paper proposes DTF, a one-shot font generation approach. Firstly, we extract the content features and style features of fonts in different dimensions. Secondly, the style and content are integrated in the high-dimensional latent space, and repeated multiple times can better combine the content details in the source image and the reference image. Finally, we perform contrastive learning between the reference image and font images with different styles. Our experimental show that our method works well few-shot font generation.