aigc3d

LAM: Create Animatable 3D Gaussian Avatars from One Image

LAM reconstructs a 3D Gaussian head avatar from a single image and supports animation and rendering across devices. It is aimed at developers building digital humans, especially interactive avatars, and requires model assets and a compatible compute environment for local use.

View on GitHub Compare
Stars
1.1k
Forks
106
Stars, last 30 days
+20

Our take

Health score

25/100

How it is scored
Activity
19
Community
41
Issues
21
Pull requests
20

+20 stars in the last 30 days

  • Fewer contributors than 82% of the projects we track

Project history

Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.

Selected snapshot

Stars

1,097

Forks

110

Change since Nov 15, 2026: 0 stars·0 forks

Overview

LAM is a Python implementation of a model that turns one input image into an animatable 3D Gaussian head avatar. It addresses the effort of creating a digital human by producing a reusable avatar representation from a single image, with animation and rendering intended to work across platforms.

The project is most relevant to graphics and avatar developers who want to experiment with image-based avatar reconstruction or connect generated avatars to interactive applications. Online demos offer a way to try it before setting up local inference.

Key Features

  • Reconstructs an animatable 3D Gaussian head from one image.
  • Provides model weights trained on VFHQ and NeRSemble datasets.
  • Includes inference scripts and a Gradio application.
  • Supports animation and rendering, with a WebGL renderer available as a related project.
  • Offers avatar export instructions for use with OpenAvatarChat.
  • Provides an audio-driven animation model through the related Audio2Expression project.
  • Includes online demos on Hugging Face and ModelScope.

Use Cases

  • Digital-human developers: Generate a starting avatar from a portrait when building a prototype or interactive experience.
  • Graphics researchers: Explore single-image reconstruction and animatable Gaussian head representations.
  • Chat application teams: Export generated avatars for integration with OpenAvatarChat.
  • Creators testing avatar pipelines: Try the hosted demos before downloading weights and configuring local inference.

Getting Started

Clone the repository and install its CUDA 12.1 dependencies on Linux:

git clone https://github.com/aigc3d/LAM.git
cd LAM
sh ./scripts/install/install_cu121.sh

See the README for CUDA 11.8 and Windows installation options, model and asset downloads, and inference instructions. Online demos are linked from the project page.

Alternatives

  • LivePortrait: LivePortrait animates a portrait from driving video or motion templates rather than reconstructing a 3D Gaussian head avatar from one image.
  • audio2photoreal: audio2photoreal generates photorealistic Codec Avatars driven by audio, rather than reconstructing a Gaussian head avatar from a single image.
ProjectLanguageLicenseStarsStatus
LAMPythonApache-2.01.1kNot checked yet
LivePortraitPythonOther19.2kQuiet
audio2photorealPythonOther2.8kArchived

Considerations

  • Local use requires downloading model weights and supporting assets separately, then installing CUDA-compatible dependencies.
  • The supplied installation scripts target CUDA 12.1 or 11.8 on Linux; Windows has a separate guide.
  • The repository describes LAM-20K trained on VFHQ and NeRSemble. It also lists a large-dataset model as a to-do item, so do not assume that variant is available.
  • Interactive chat and audio-driven animation involve related projects and integration steps rather than being a single built-in, standalone feature.

Found this useful?

Share it with someone who would like LAM.

Comparisons