GASP: Gaussian Avatars with Synthetic Priors

Figure 1 . We propose GASP, a novel model for creating photorealistic, real-time, animatable, 360 • avatars from easily-captured data. We train a generative prior model of Gaussian Avatars on Synthetic data. The prior allows our model to be fit using a single image or a short video with the prior accounting for the unseen views. This lets users create their avatar with only a webcam or smartphone.