DreamHuman is a method developed by Google Research that generates realistic, animatable 3D human avatar models solely from textual descriptions. It aims to address the limitations of recent text-to-3D methods, such as limited control and spatial resolution, inability to produce animated 3D human models, and challenges in maintaining anthropometric consistency for complex structures like humans. DreamHuman combines large text-to-image synthesis models, neural radiance fields, and statistical human body models in a unique modeling and optimization framework. This allows it to generate dynamic 3D human avatars with high-quality textures and learned, instance-specific, surface deformations.
Publication date: June 15, 2023
Project Page: https://dream-human.github.io
Paper: https://arxiv.org/pdf/2306.09329.pdf
