Researchers have developed Generative Relightable Avatars (GRA), a novel method for creating photorealistic, free-view renderings of full-body humans that can be relit under different environment maps. GRA employs a hybrid approach, first optimizing material parameters in UV-space to render a coarse appearance under a target HDR environment map. This is then refined by a feed-forward model to capture pose-dependent dynamics and illumination effects. Finally, a fine-tuned video-to-video diffusion model transforms these renderings into detailed, temporally coherent videos, with an error-recycling strategy for generating longer sequences. AI
IMPACT This research advances generative models for 3D avatar creation, potentially impacting virtual reality and content generation.
RANK_REASON The cluster describes a new research paper detailing a novel method for generating relightable avatars. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
