HanDiffuser: Text-to-Image Generation With Realistic Hand Appearances

TL;DR

HanDiffuser is a novel diffusion-based model that significantly improves the realism of generated hands in text-to-image synthesis by incorporating detailed hand representations and a two-stage generative process.

Contribution

The paper introduces HanDiffuser, a new architecture that explicitly models hand parameters to enhance hand realism in generated images, addressing a key challenge in text-to-image models.

Findings

Achieves high-quality, realistic hand generation in images.

Outperforms existing models in quantitative and qualitative evaluations.

User studies confirm improved hand realism.

Abstract

Text-to-image generative models can generate high-quality humans, but realism is lost when generating hands. Common artifacts include irregular hand poses, shapes, incorrect numbers of fingers, and physically implausible finger orientations. To generate images with realistic hands, we propose a novel diffusion-based architecture called HanDiffuser that achieves realism by injecting hand embeddings in the generative process. HanDiffuser consists of two components: a Text-to-Hand-Params diffusion model to generate SMPL-Body and MANO-Hand parameters from input text prompts, and a Text-Guided Hand-Params-to-Image diffusion model to synthesize images by conditioning on the prompts and hand parameters generated by the previous component. We incorporate multiple aspects of hand representation, including 3D shapes and joint-level finger positions, orientations and articulations, for robust learning and reliable performance during inference. We conduct extensive quantitative and qualitative experiments and perform user studies to demonstrate the efficacy of our method in generating images with high-quality hands.