Large-scale Codec Avatars: The Unreasonable Effectiveness of Large-scale Avatar Pretraining

• URL: http://arxiv.org/abs/2604.02320v2
• Date: Tue, 07 Apr 2026 17:25:24 GMT
• Title: Large-scale Codec Avatars: The Unreasonable Effectiveness of Large-scale Avatar Pretraining
• Abstract summary: Large-Scale Codec Avatars (LCA) is a high-fidelity, full-body 3D avatar model that generalizes to world-scale populations in a feedforward manner.<n>LCA generalizes across hair styles, clothing, and demographics while providing precise, fine-grained facial expressions and finger-level articulation control.
• Score: 62.501929209687056
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: High-quality 3D avatar modeling faces a critical trade-off between fidelity and generalization. On the one hand, multi-view studio data enables high-fidelity modeling of humans with precise control over expressions and poses, but it struggles to generalize to real-world data due to limited scale and the domain gap between the studio environment and the real world. On the other hand, recent large-scale avatar models trained on millions of in-the-wild samples show promise for generalization across a wide range of identities, yet the resulting avatars are often of low-quality due to inherent 3D ambiguities. To address this, we present Large-Scale Codec Avatars (LCA), a high-fidelity, full-body 3D avatar model that generalizes to world-scale populations in a feedforward manner, enabling efficient inference. Inspired by the success of large language models and vision foundation models, we present, for the first time, a pre/post-training paradigm for 3D avatar modeling at scale: we pretrain on 1M in-the-wild videos to learn broad priors over appearance and geometry, then post-train on high-quality curated data to enhance expressivity and fidelity. LCA generalizes across hair styles, clothing, and demographics while providing precise, fine-grained facial expressions and finger-level articulation control, with strong identity preservation. Notably, we observe emergent generalization to relightability and loose garment support to unconstrained inputs, and zero-shot robustness to stylized imagery, despite the absence of direct supervision.

Related papers

• Autoregressive Appearance Prediction for 3D Gaussian Avatars [67.37159422872219]
  We propose a 3D Splatting avatar model with a spatial backbone conditioned on both pose and an appearance latent.<n>The latent is learned during training by an encoder, yielding a compact representation that improves reconstruction quality.<n>Our method delivers a robust and practical path to high-fidelity, stable avatar driving.
  arXiv  Detail & Related papers  (2026-04-01T14:07:14Z)
• FRESA: Feedforward Reconstruction of Personalized Skinned Avatars from Few Images [74.86864398919467]
  We present a novel method for reconstructing personalized 3D human avatars with realistic animation from only a few images.<n>We learn a universal prior from over a thousand clothed humans to achieve instant feedforward generation and zero-shot generalization.<n>Our method generates more authentic reconstruction and animation than state-of-the-arts, and can be directly generalized to inputs from casually taken phone photos.
  arXiv  Detail & Related papers  (2025-03-24T23:20:47Z)
• Multimodal Generation of Animatable 3D Human Models with AvatarForge [67.31920821192323]
  AvatarForge is a framework for generating animatable 3D human avatars from text or image inputs using AI-driven procedural generation.<n>Our evaluations show that AvatarForge outperforms state-of-the-art methods in both text- and image-to-avatar generation.
  arXiv  Detail & Related papers  (2025-03-11T08:29:18Z)
• XAGen: 3D Expressive Human Avatars Generation [76.69560679209171]
  XAGen is the first 3D generative model for human avatars capable of expressive control over body, face, and hands. We propose a multi-part rendering technique that disentangles the synthesis of body, face, and hands. Experiments show that XAGen surpasses state-of-the-art methods in terms of realism, diversity, and expressive control abilities.
  arXiv  Detail & Related papers  (2023-11-22T18:30:42Z)
• Text-Conditional Contextualized Avatars For Zero-Shot Personalization [47.85747039373798]
  We propose a pipeline that enables personalization of image generation with avatars capturing a user's identity in a delightful way. Our pipeline is zero-shot, avatar texture and style agnostic, and does not require training on the avatar at all. We show, for the first time, how to leverage large-scale image datasets to learn human 3D pose parameters.
  arXiv  Detail & Related papers  (2023-04-14T22:00:44Z)
• DreamAvatar: Text-and-Shape Guided 3D Human Avatar Generation via Diffusion Models [55.71306021041785]
  We present DreamAvatar, a text-and-shape guided framework for generating high-quality 3D human avatars. We leverage the SMPL model to provide shape and pose guidance for the generation. We also jointly optimize the losses computed from the full body and from the zoomed-in 3D head to alleviate the common multi-face ''Janus'' problem.
  arXiv  Detail & Related papers  (2023-04-03T12:11:51Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.