Reproduce “FaceRiff: Data-Efficient Unified Text-to-3D Face Asset Synthesis via Geometric Conditioning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Elly Akhoundi, Mathieu Lamarre, Thibault Goyallon, Patrick Anderson, Étienne Danvoye
FaceRiff generates face assets from text, adapting pretrained SDXL with LoRA and normal-map conditioning to produce a mesh, albedo, and displacement in a shared UV layout. Canonical normals guide mesh generation, while predicted-mesh normals guide color and relief. With matched ground-truth meshes, experts judged its renderings real in 59.4% of examples, versus 8.9% for DreamFace; this result concerns appearance, not the full pipeline.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.