Apple’s GHARP enables real-time 3D head animation from sparse images using Gaussian splats
New method decouples identity modeling from expression-driven animation, achieving state-of-the-art results on Ava-256 while running up to 13× faster on an A100 GPU.
1 source · cross-referenced
- Apple researchers propose GHARP, a two-stage method for real-time 3D head animation from a few input images and a driving expression signal.
- The identity stage builds a subject-specific Gaussian representation offline, while the animation stage predicts expression-dependent residuals at runtime.
- GHARP achieves state-of-the-art quality on the Ava-256 benchmark and runs up to 13× faster on an A100 GPU with 8× fewer Gaussians.
- A body alignment network addresses ambiguity in body pose and clothing by aligning the target body with reference images.
Apple’s Machine Learning Research team introduced GHARP (Real-time Gaussian Head Animation from Large-scale Reconstruction Prior), a method that animates 3D human heads in real time from a few input images of a subject and a driving expression signal.
The approach decouples the problem into two stages: an identity stage that builds a representation of the subject’s geometry and appearance offline, and an animation stage that predicts expression-dependent residuals on top of the identity representation at runtime. This separation allows the identity stage to be computationally expensive while keeping the animation stage lightweight, optimized for mobile devices.
GHARP operates within a semantically structured latent space of a pretrained reconstruction model, where expression changes remain spatially contained. This design enables efficient residual prediction and allows fusion of multiple input views into a compact, fixed-size canonical Gaussian representation.
The authors note a common limitation of expression-driven avatar methods: expression codes typically describe only the face, omitting body pose and clothing. This omission introduces ambiguity, causing the animation network to average over conflicting body appearances and produce blur or temporal flicker. To address this, GHARP includes a body alignment network that learns to align the person’s body in the target image with the input reference images, removing ambiguity from the training signal.
On the Ava-256 benchmark, GHARP achieves state-of-the-art quality while running up to 13× faster on an A100 GPU and using 8× fewer Gaussians compared to prior methods.
- Nov 11, 2026 · The Verge — AI
Google DeepMind unveils AlphaGenome 2 to predict genetic variant effects on cell types
Trust79 - Nov 11, 2026 · Google DeepMind — Blog
Google DeepMind unveils AlphaGenome 2 to predict DNA variant effects in specific cell types
Trust84 - Nov 11, 2026 · arXiv cs.AI
Researchers demonstrate auditable human-agent teams for autonomous synchrotron X-ray experiments
Trust79