The model extends a NeRF into a generative neural model that can synthesize multiview-consistent images.

Have a look at "MoRF: Morphable Radiance Field for Multiview Neural Head Modeling" – a new method that extends a neural radiance field (NeRF) into a generative neural model that can realistically synthesize images of complete human heads with variable and controllable identities.

Simply put, MoRF combines two heads and makes a new one, which has features of both "parent" heads. Something like Artbreeder but in 3D. The model provides realistic and consistent rendering under novel viewpoints.

MoRF was trained by leveraging a high-quality database of multiview portrait images of several people captured in the studio with polarization-based separation of diffuse and specular reflection. The researchers believe MoRF is a strong new step towards 3D morphable neural head modeling.

If you'd like to know more, check out the research paper here.