China Open Source Digital Human Projects
Open-source digital-human projects from China, grouped by what they do: talking heads and lip-sync, portrait and body animation, character and custom video, 3D avatars, conversational and live-streaming systems, production platforms, SDKs and voice tools, with a closing list of notes and references. Each section runs A–Z; every entry gives the project, its maker and its code, then what it does. Copies, mirrors and forks of a project on GitHub or Gitee are listed with its code links, as are the maker’s own ModelScope releases, which are easier to reach from inside China. Research projects that have published a paper and demo page but no code are marked “project page only”.
Checked October 2026. More Chinese-language projects turn up in a search for 数字人 on Gitee (search.gitee.com) and on ModelScope (modelscope.cn).
Talking heads and lip-sync
ACTalker · HKUST with Tencent and Tsinghua University · GitHub
Generates talking-head video controlled by audio, by facial motion or by both at once, using a video diffusion model with selective state-space control; an ICCV 2025 paper.
AniPortrait · Tencent (Tencent Games Zhiji) · GitHub
Audio-driven photorealistic portrait animation from one reference image, first predicting facial landmarks from the audio and then rendering video with a diffusion model.
AniTalker · Shanghai Jiao Tong University (X-LANCE lab) · GitHub
Animates varied, lively talking faces from one portrait and an audio track, using a facial-motion encoding kept separate from identity; an ACM MM 2024 paper.
DAWN · University of Science and Technology of China with iFLYTEK Research · GitHub
Talking-head video generated by a non-autoregressive diffusion model that produces all frames at once, for videos of any length.
DH_live · kleinlee · GitHub · Gitee copy (aetherbyte) · MatesX
A fast 2D video digital human whose mini version runs in real time in a phone’s web browser with no GPU; its successor MatesX adds memory, expressions and actions on Windows, macOS, iOS, Android and mini programs.
DICE-Talk · Fudan University with Tencent (YouTu Lab) · GitHub
Emotional talking-head generation from a portrait and audio that keeps identity apart from emotion, giving vivid and varied expressions; an ACM MM 2025 paper.
digital-wavlipgfp · Filifun · GitHub
Makes lip-synced short videos from a source video and new audio, syncing the mouth with Wav2Lip and then sharpening the face with GFP-GAN, as a stand-in for HeyGen.
DINet · NetEase · GitHub
Visual dubbing of high-resolution face video: a deformation-inpainting network re-renders the mouth to match new audio, with training code included.
DisentTalk · Institute of Information Engineering, Chinese Academy of Sciences · project page only
Talking-face generation built for accurate lip-sync across languages, using a diffusion model that separates facial semantics and handles space and time together; an ICME 2025 paper.
Ditto · Ant Group · GitHub
Controllable real-time talking-head synthesis by diffusion in a motion space; an ACM MM 2025 paper.
Double_Talker · lyyong · Gitee
Shows two digital humans speaking on one screen, driven in a loop by Wav2Lip, with an upscaling step afterwards.
DreamTalk · Tsinghua University with Alibaba Group · GitHub
Expressive talking-head generation with diffusion models that holds up across languages, noisy audio and unusual portraits.
DualTalk · Renmin University of China with Ant Group · GitHub
A 3D talking-head model for two-person conversations that switches between speaking and listening and adds non-verbal feedback; a CVPR 2025 paper.
ER-NeRF · Beihang University · GitHub
High-fidelity talking-portrait synthesis with region-aware neural radiance fields, trained on a short video of one speaker; an ICCV 2023 paper.
GaussianTalker · Alibaba Group (Taobao (China) Software) · project page only
Speaker-specific talking-head synthesis with 3D Gaussian splatting; an ACM MM 2024 paper from the team behind TaoAvatar.
GeneFace++ · Zhejiang University · GitHub
Real-time 3D talking-face generation for a particular speaker, with stable, well-generalised lip-sync.
Hallo · Fudan University with Baidu · GitHub Hallo · Hallo2 · Hallo3
Audio-driven portrait animation in three generations: Hallo with hierarchical audio-visual synthesis, Hallo2 for long-duration, high-resolution video (ICLR 2025), and Hallo3 with a video diffusion transformer for highly dynamic scenes (CVPR 2025).
JoyGen · JD.com with the University of Hong Kong · GitHub
Re-edits the lips in an existing talking-face video to match new audio, in two stages: audio-driven lip motion, then image synthesis guided by 3D depth.
JoyHallo · JD Health · GitHub
A digital-human video model for Mandarin, built on Hallo and trained on a Mandarin video dataset recorded by JD Health staff.
JoyVASA · JD Health with Zhejiang University · GitHub
Audio-driven portrait animation that separates facial dynamics from the static face, and can animate animal faces as well as human ones.
LatentSync · ByteDance with Beijing Jiaotong University · GitHub
Lip-sync that maps audio straight to video frames in Stable Diffusion’s latent space; version 1.6 (June 2025) is trained on 512×512 video to cut blur.
LiteAvatar · Alibaba Group (HumanAIGC) · GitHub · ModelScope avatar gallery
An audio-to-face model for real-time 2D chat avatars that runs at 30 frames a second on a CPU alone, and on mobile devices.
LLIA · Meituan · project page only
Low-latency interactive avatars: real-time audio-driven portrait video generated with diffusion models.
LstmSync · oneCodeSuperman · GitHub
A general lip-sync model that needs no per-person training; version 2 runs on low video memory with audio of any length.
MEMO · Skywork AI (Kunlun Tech) with the National University of Singapore · GitHub · ModelScope
Memory-guided, emotion-aware diffusion for expressive talking video from one image and audio.
MimicTalk · Zhejiang University · GitHub
Adapts a personalised, expressive 3D talking face to a new speaker in minutes; a NeurIPS 2024 paper.
MuseTalk · TME Lyra Lab · GitHub
Real-time lip-sync that inpaints the mouth region of existing video in latent space, at 30 frames a second or more on a V100 GPU.
OmniSync · Renmin University of China with Kuaishou (Kling team) · project page only
Lip-sync with diffusion transformers that holds across occlusions, head poses and stylised faces.
Real3D-Portrait · Zhejiang University · GitHub
One-shot realistic 3D talking portraits from a single image, driven by audio or video; an ICLR 2024 spotlight paper.
SadTalker · Xi’an Jiaotong University with Tencent AI Lab and Ant Group · GitHub
Animates a talking face from one image and an audio track by generating realistic 3D motion coefficients; a CVPR 2023 paper.
sadtalker-kaggle · yuxiaolinglan · GitHub
SadTalker, which animates a talking face from one image, packaged to run on Kaggle, with face-restoration and head-movement options added to its Gradio page.
Sonic · Tencent with Zhejiang University · GitHub · ComfyUI node (smthemex)
Speech-driven portrait animation that reads the whole audio track to drive lips, head and expression; a CVPR 2025 paper.
SoulX-FlashHead · Soul App · GitHub
A 1.3-billion-parameter model for high-fidelity streaming portrait video in real time and of unlimited length.
SyncTalk · Renmin University of China with Beijing University of Posts and Telecommunications, the Chinese Academy of Sciences, Tsinghua University and Psyche AI · GitHub
NeRF talking heads with tighter synchronisation of lips, expression and head pose; a CVPR 2024 paper.
TalkingGaussian · Beihang University · GitHub
3D talking-head synthesis with Gaussian splatting that keeps facial structure intact while the mouth moves; an ECCV 2024 paper.
Ultralight-Digital-Human · anliyuan · GitHub · FeatherTalk
A very lightweight lip-sync digital-human model that runs in real time on mobile devices; its C++ streaming successor is FeatherTalk.
V-Express · Tencent AI Lab · GitHub
Talking-head video from a reference image, an audio track and a sequence of facial keypoint images.
xuni8 · zhangbo2008 · GitHub
A developer’s working code for fine-tuning Wav2Lip on Chinese video, with a sync classifier, CodeFormer upscaling and a web front end.
Portrait and body animation
Animate-X · Ant Group with Alibaba Group (Tongyi Lab) · GitHub · Animate-X++
Animates any character image, cartoon and non-human figures included, from a pose sequence; an ICLR 2025 paper, followed by Animate-X++.
AnimateAnyone · Alibaba Group (Institute for Intelligent Computing) · project page only
Consistent, controllable animation of a character from one image and a pose sequence.
AptAvatar · Alibaba Group (Taobao Live AIGC team) · GitHub · ModelScope
Fast, vivid long-form audio-driven avatar video aimed at production use, built on a distilled few-step video model.
Champ · Nanjing University with Fudan University and Alibaba Group · GitHub
Human image animation guided by a 3D parametric body model, for controllable, consistent motion; an ECCV 2024 paper.
ChatAnyone · Alibaba Group (Tongyi Lab) · project page only
Real-time stylised portrait video from one image and an audio track, from the head alone up to the upper body, with expression and style control.
CoDance · The University of Hong Kong with Ant Group, Huazhong University of Science and Technology and Alibaba Group (Tongyi Lab) · GitHub
Animates any number of characters in any arrangement within one image, without training a model for each character.
DreamActor-M1 · ByteDance · project page only
Animates one reference photo into video that copies a driving video’s facial expression and body movement, using a diffusion transformer with hybrid guidance; an ICCV 2025 paper.
EchoMimic · Ant Group (Alipay) · GitHub V1 · V2 · V3 · V2 page · ModelScope
Audio-driven portrait animation: V1 animates a face from audio and editable facial landmarks, V2 extends it to half-body animation with hand poses, and V3, at 1.3 billion parameters, unifies several human-animation tasks in one model.
EMO · Alibaba Group (Institute for Intelligent Computing) · project page only
Expressive portrait video, talking or singing, generated straight from one image and audio; an ECCV 2024 paper.
FantasyPortrait · Alibaba Group (Amap) · GitHub
Expression-driven portrait animation for one or several characters, using diffusion transformers.
FantasyTalking · Alibaba Group (Amap) with Beijing University of Posts and Telecommunications · GitHub · ModelScope
Turns one portrait and an audio track into a talking video with matching expressions and body movement; an ACM MM 2025 paper.
FlowAct-R1 · ByteDance · project page only
Real-time humanoid video generation that responds to a live conversation instead of being rendered in advance.
Follow-Your-Emoji · HKUST · GitHub
Portrait animation driven by expression landmarks, including exaggerated expressions, with a faster later version; SIGGRAPH Asia 2024 and IJCV 2025.
HumanDiT · Zhejiang University with ByteDance · project page only
Long human-motion videos from one image, guided by pose, with a diffusion transformer that keeps hands and faces detailed at varying resolutions.
HunyuanPortrait · Tencent (Hunyuan) · GitHub
Portrait animation from a driving video, using implicit condition control; a CVPR 2025 paper.
HunyuanVideo-Avatar · Tencent (Hunyuan) with TME Lyra Lab · GitHub · ModelScope
Audio-driven animation of one or several characters from images, with emotion matched to the audio.
InfiniteTalk · Meituan · GitHub
Audio-driven “sparse-frame video dubbing”: re-animates the lips, head, body and expression in an existing video to fit new audio, at any length.
JoyAI-Echo · JD.com · GitHub
Long-form audio-visual generation that keeps each character’s appearance and voice consistent across multi-shot videos of up to five minutes, with a memory bank and a director agent; aimed at digital-human livestreaming, virtual IP stories, marketing and education.
Live Avatar · Alibaba Group with University of Science and Technology of China · GitHub · ModelScope
Streams audio-driven avatar video in real time and without a length limit; an ECCV 2026 spotlight paper.
LivePortrait · Kuaishou · GitHub
Animates a still portrait to follow the motion of a driving video, with stitching and retargeting controls for the eyes and lips, fast enough for real-time use.
LongCat-Video-Avatar · Meituan · GitHub · ModelScope
Audio-driven avatar video from one portrait and an audio track, with natural movement during silences; version 1.5 improves lip-sync, multi-person interaction and long-video stability and cuts inference to eight steps.
MimicMotion · Tencent with Shanghai Jiao Tong University · GitHub
High-quality human motion video from a pose sequence, with confidence-aware pose guidance.
MiniMates · kleinlee · GitHub
A lightweight image-driven digital-human algorithm, which its author measures at 10 to 100 times faster than LivePortrait, EchoMimic and MuseTalk, with voice-driven and expression-driven modes that run on an ordinary computer.
MultiTalk · Meituan · GitHub
Audio-driven video of several people in conversation, each bound to their own audio track; a NeurIPS 2025 paper.
MusePose · TME Lyra Lab · GitHub
A pose-driven image-to-video framework for virtual humans.
MuseV · TME Lyra Lab · GitHub
Virtual-human video generation of unlimited length with visual-conditioned parallel denoising.
OmniAvatar · Zhejiang University with Alibaba Group · GitHub · ModelScope
Audio-driven full-body avatar video built on the Wan video model, with body movement adapted to the audio.
OmniHuman-1 · ByteDance · project page only
Generates human video, from portrait to full body, out of one image plus motion signals such as audio, video or both.
SkyReels-A1 · Skywork AI (Kunlun Tech) · GitHub
Expressive portrait animation with video diffusion transformers.
SkyReels-V3 · Skywork AI (Kunlun Tech) · GitHub · ModelScope talking-avatar model
A multimodal video model with three capabilities in one architecture: multi-subject video from reference images, audio-driven video (talking avatars) and video-to-video.
SoulX-FlashTalk · Soul App · GitHub
A 14-billion-parameter audio-driven avatar model that streams in real time, starting in 0.87 seconds and running at 32 frames a second on an 8×H800 node.
SoulX-LiveAct · Soul App · GitHub
Real-time streaming human animation that stays stable for hours, keeping identity and lip-sync consistent; runs at 20 frames a second at 512×512 on two H100 or H200 GPUs.
StableAnimator · Fudan University with Microsoft Research Asia and Huya · GitHub
Identity-preserving human image animation from a pose sequence, end to end; a CVPR 2025 paper.
StableAvatar · Fudan University with Microsoft Research Asia, Xi’an Jiaotong University and Tencent · GitHub · ModelScope
Audio-driven avatar video of unlimited length from an end-to-end video diffusion transformer.
UniAnimate · Alibaba Group (Tongyi) with the University of Science and Technology of China · GitHub · UniAnimate-DiT
Consistent human image animation with unified video diffusion models; UniAnimate-DiT scales it up on a large video diffusion transformer.
VividTalk · Alibaba Group (HumanAIGC) · project page only
One-shot audio-driven talking-head generation using a hybrid 3D prior.
Wan2.2 · Alibaba Group (Tongyi Wan) · GitHub
Open video generation models, including Wan2.2-S2V, which makes audio-driven video of speaking characters, and Wan2.2-Animate, which animates or replaces a character to follow a driving video.
X-Portrait · ByteDance · GitHub
Expressive portrait animation from a driving video, with hierarchical motion attention; a SIGGRAPH 2024 paper.
Character images and custom video
FaceChain · Alibaba Group (ModelScope) · GitHub · Gitee mirror
Builds a personal digital twin from a few photos and generates portraits of it in chosen styles.
HunyuanCustom · Tencent (Hunyuan) · GitHub
Generates video that keeps a chosen subject consistent, conditioned on text, images, audio or video.
InstantCharacter · Tencent (Hunyuan) with InstantX · GitHub
Places any character from one reference image into new images, using a diffusion transformer with an adapter.
InstantID · InstantX with Peking University · GitHub
Identity-preserving image generation from a single face photo in seconds, without fine-tuning.
PhotoMaker · Tencent (ARC Lab) with Nankai University · GitHub
Generates realistic photos of a person in new styles and settings from a few reference photos; a CVPR 2024 paper.
SkyReels-A2 · Skywork AI (Kunlun Tech) · GitHub
Composes a video from several reference elements, such as characters, objects and backgrounds, with video diffusion transformers.
3D avatars
AniGen · VAST with The University of Hong Kong · GitHub
Generates animatable 3D assets, skeleton and skinning included, from a single image; a SIGGRAPH 2026 paper.
AniGS · Alibaba Group (Tongyi Lab) · GitHub
An animatable Gaussian avatar from a single image; a CVPR 2025 paper.
AvatarArtist · HKUST with Ant Group · GitHub
Creates an animatable 4D avatar from one open-domain image, photographic or stylised, keeping its identity and style; a CVPR 2025 paper.
CharacterGen · Tsinghua University with VAST · GitHub
Generates a 3D character from a single image, putting it into a canonical pose across multiple views; a SIGGRAPH 2024 paper.
CodeTalker · The Chinese University of Hong Kong with Tencent AI Lab · GitHub
Speech-driven 3D facial animation from a learned codebook of discrete facial motions; a CVPR 2023 paper.
DiffPortrait360 · University of Southern California, MBZUAI and Pinscreen with ByteDance and CUHK-Shenzhen · GitHub
Synthesises a consistent 360-degree view of a whole head, back included, from one portrait without fine-tuning; a CVPR 2025 paper.
DiffPoseTalk · Tsinghua University · GitHub
Speech-driven 3D facial animation and head pose in a chosen speaking style, generated by diffusion, with the TFHP dataset released alongside.
FaceFormer · The University of Hong Kong · GitHub
Speech-driven 3D facial animation with transformers; a CVPR 2022 paper.
FACEGOOD Audio2Face · FACEGOOD · GitHub
Drives a 3D face’s expressions from voice in real time, released to lower the cost of building AI digital humans.
GaussianIP · Beihang University with Shanghai Artificial Intelligence Laboratory · GitHub
Generates identity-preserving 3D human avatars from a text prompt and a reference image in about 40 minutes; a CVPR 2025 paper.
GUAVA · Tsinghua University (Shenzhen) with IDEA · GitHub
Builds an animatable 3D upper-body Gaussian avatar from one image in under a second, then animates it in real time; an ICCV 2025 paper.
HRAvatar · Tsinghua University (Shenzhen) with IDEA · GitHub
Reconstructs high-quality, relightable 3D head avatars from ordinary single-camera video using Gaussian splatting; a CVPR 2025 paper.
HumanGaussian · The Chinese University of Hong Kong with Tencent AI Lab · GitHub
Generates 3D humans from a text prompt with Gaussian splatting; a CVPR 2024 highlight.
Hunyuan3D-2 · Tencent · GitHub
Generates high-resolution textured 3D assets from text or images; a general 3D model rather than an avatar model.
HY-Motion 1.0 · Tencent (Hunyuan) · GitHub
Generates skeleton-based 3D character animation from text, using a diffusion transformer with flow matching.
LAM · Alibaba Group (Tongyi Lab) · GitHub · ModelScope demo
Large Avatar Model: an animatable Gaussian head avatar from one image in seconds, rendered across platforms, with a low-latency SDK for real-time chat avatars; a SIGGRAPH 2025 paper.
LAM Audio2Expression · Alibaba Group (Tongyi Lab) · GitHub
Generates ARKit facial expressions from audio in real time, to drive LAM head avatars.
LHM · Alibaba Group (Tongyi Lab) · GitHub · Gitee copy (francis-v01)
Reconstructs an animatable 3D human from a single image in seconds.
Make-A-Character · Alibaba Group (Institute for Intelligent Computing) · project page only · ModelScope demo
Generates realistic, animatable 3D characters from a text description within minutes; the ModelScope demo turns the character into a presenter video from a script and a chosen voice.
MNN-TaoAvatar · Alibaba Group · GitHub · Gitee mirror · ModelScope
An Android app that runs a 3D talking avatar entirely on the phone, with speech recognition, a language model, speech synthesis and avatar rendering, built on Alibaba’s MNN inference engine and the TaoAvatar paper.
Multiview-Avatar · Alibaba Group (Tongyi Lab) · GitHub · ModelScope
Generates consistent multi-view images of a character from a text description, built on Stable Diffusion 1.5 with ControlNet.
PSHuman · Tsinghua University and Southeast University with VAST · GitHub
Reconstructs a detailed clothed 3D human from a single image in about a minute, using multi-view diffusion.
RGBAvatar · Zhejiang University · GitHub
Builds photoreal head avatars from video in real time, using a reduced set of Gaussian blendshapes; a CVPR 2025 highlight.
SEGA · Tsinghua University (Shenzhen) with ByteDance · project page only
A drivable 3D Gaussian head avatar from a single image.
SentiAvatar · SentiPulse with Renmin University of China (Gaoling School of Artificial Intelligence) and Inner Mongolia University · GitHub
An interactive 3D digital-human framework, released with the 3D character SUSU and the SuSuInterActs motion dataset.
TexTalker · Shanghai Jiao Tong University · GitHub
Generates 3D talking avatars from speech, with mesh motion and dynamic facial texture produced together, trained on the authors’ TexTalk4D dataset; a CVPR 2025 paper.
THGS · University of Science and Technology of China with Anhui University and the Hefei Comprehensive National Science Center · GitHub (dataset only)
Learns a 3D Gaussian-splatting talking human avatar from about a minute of single-camera video, with control over expression, body joints and camera, rendering at over 150 frames a second in a web viewer; a Computer Graphics Forum 2025 paper.
UniRig · Tsinghua University with VAST · GitHub
Automatic skeleton rigging and skinning for many kinds of 3D model, from humans to animals and fictional characters, trained on the Rig-XL dataset; a SIGGRAPH 2025 paper.
UniTalker · SenseTime · GitHub
A unified audio-driven 3D facial animation model trained across many datasets with different mesh formats, released with the data.
Zero-1-to-A · Zhejiang University · project page only
An animatable head avatar from one image using video diffusion, with no per-subject tuning; a CVPR 2025 paper.
Conversational digital humans
account-digitalPlayer-Unity3D · MingZheGe · GitHub
A digital human that guides a customer through opening an account, with a Vue front end and a Java back end.
ai-character-hub · java_lovely · Gitee
A Spring Boot platform for creating digital characters, covering model training, speech synthesis and interaction design.
AIAvatar · shibing624 · GitHub
A real-time interactive digital human that holds audio-video synchronised conversations, with Wav2Lip lip-sync.
any4any · eogee · Gitee
An enterprise multimodal AI platform combining LLM chat, a digital-human system, natural-language database queries, speech and a knowledge base behind an OpenAI-compatible API.
Awesome Digital Human · wan-h · GitHub · Gitee copy (baixjcy)
A lightweight framework, running on two CPU cores and 2 GB of memory, that links a Live2D character to modular ASR, LLM, TTS and agent services, including Dify, FastGPT and Coze; the community site light4ai.com is built on it.
city-exhibition-guide · xtdexw · GitHub
An AI guide for a city exhibition hall, with a 3D digital human on a large screen, built on MoFa’s Xingyun SDK for a MoFa hackathon.
CyberVerse · Lynpoint · GitHub
A self-hosted real-time digital-human agent platform: voice-first agents over WebRTC with persona memory, tools and knowledge retrieval, and optional digital-human video from a single photo using the SoulX-FlashHead and SoulX-LiveAct models.
digital-human-realtime · Yuanyue Technology · GitHub
A Windows real-time conversational digital human for local deployment in customer service, shopping guidance and government services, with Wav2Lip lip-sync and switchable avatars.
Distilly (formerly colleague-skill) · titanwings · GitHub · similar projects: nuwa-skill, yourself-skill, ex-skill, boss-skill, parents-skills, Xue-Feng-Skill
Distils a person’s chat records, documents and work style into a reusable AI skill that answers the way they would; its 2026 release as colleague.skill (同事.skill) started a wave of similar ‘digital double’ skills of ex-partners, bosses, parents and public figures.
EdgePersona · zc-maker · GitHub
A fully offline digital human that runs in real time on a laptop with an NVIDIA 3060, with voice conversation, expressions and body motion, and characters configured in YAML.
Edubot · caixukun-jinitaimei · GitHub
An education system built on Linly-Talker, with lecture summaries, digital-human dialogue and chatbot dialogue, deployable on AutoDL.
Fay · xszyou · GitHub · Android · UE5 · Gitee copies: OldBlack, YFKael · forks and copies: ArtificialZeng, ascanzen, flyarong, lili3533, muttofan, shuxiaokai
An agent framework that connects digital humans (2.5D, 3D, mobile, PC and web) and LLMs (OpenAI-compatible, DeepSeek) to business systems, with swappable ASR and TTS, offline use, and interfaces for microcontrollers, apps and websites; an Android connector and an Unreal Engine 5 MetaHuman project go with it.
general_digital_human_system · HansonJames · GitHub
A conversational digital human using Microsoft Azure avatar rendering over WebRTC, with GPT dialogue, knowledge-base retrieval and emotion sync, built on FastAPI and LangChain.
GMTalker · Guangming Laboratory (Guangdong Provincial Laboratory of Artificial Intelligence and Digital Economy, Shenzhen) · GitHub
An interactive 3D digital human combining speech recognition, speech synthesis, language understanding and lip animation, deployable fully offline on Windows, Linux and Android.
KinEcho · luobo_lin · Gitee
A companion system for older people, including those with disabilities or dementia: a digital human for daily care, emotional support, medication reminders and health management, which also plays memories uploaded by family at suitable moments.
Linly-Talker · Kedreamix · GitHub · Linly-Talker-Stream, full-duplex real-time version · Gitee copies: 910024445, zhitong-zy · ModelScope
A conversational digital-human system that combines LLMs, speech recognition, speech synthesis with GPT-SoVITS voice cloning, and talking-face models including MuseTalk and NeRF methods, behind a web UI.
live2d (数字人AI项目) · kingdess · Gitee
A low-cost web digital human: a Live2D character in the browser connected to an LLM assistant through a small Python server.
live2dSpeek · lyz1810 · GitHub
A Live2D digital human in the browser that speaks text through edge-tts.
LiveTalking · lipku · GitHub · Gitee · Gitee copies: 1120855315, TF-boy · LiveTalking-v15 fork (brucezhao), adding MuseTalk 1.5, local speech recognition and Coze workflows
A real-time interactive streaming digital-human engine with synchronised audio and video, supporting ER-NeRF, MuseTalk, Wav2Lip and Ultralight models, voice cloning, interruption and WebRTC or virtual-camera output; formerly named metahuman-stream.
Metaman · DivineConcerto · GitHub
A Unity digital human whose Flask server turns the user’s speech into text, gets a GPT reply and sends it back to Unity.
my-neuro · morettt · GitHub
An AI desktop companion with a customisable Live2D character, voice conversation with about one-second replies, long-term memory, visual recognition, voice cloning through GPT-SoVITS and Bilibili live streaming.
N.E.K.O · Project N.E.K.O · GitHub
An AI companion character with real-time voice, text and vision, persistent memory, Live2D, VRM or MMD avatars, proactive conversation and agent tools.
OpenAvatarChat · Alibaba Group (Tongyi HumanAIGC) · GitHub · WebRTC component · Gitee copy (hesenwei)
A modular interactive digital-human chat that runs on a single PC, with swappable ASR, LLM, TTS and avatar modules (LiteAvatar, LAM, MuseTalk, FlashHead) and an average response time of 2.2 seconds; its WebRTC component is a fork of FastRTC.
PaddleAvatar · Kedreamix · GitHub
Turns your image, audio and video into a talking digital-human video with Baidu’s PaddlePaddle toolkits, with room to add dialogue on top.
sysu-digital-man · Fucloud233 · GitHub
A 24-hour campus guide made for Sun Yat-sen University’s centenary creative contest, built on Fay with a reworked LLM layer and a vector database.
Unity Digital Human · LKZMuZiLi · GitHub · Gitee · Gitee copy (xxhlq)
A Unity (URP) real-time 3D digital human that chains streaming speech recognition, any LLM API and speech synthesis with lip-sync, and builds for all platforms.
VideoChat · Henry-23 · GitHub · fork (hwhw97)
A real-time voice-interactive digital human, either end to end (multimodal LLM to talking head) or cascaded (ASR, LLM, TTS, talking head), with a custom avatar, voice cloning and a first response in as little as 3 seconds.
VirtualWife · yakami129 · GitHub · fork (laixiao)
A virtual companion character with long- and short-term memory, VRM character models, switchable LLMs including local models through Ollama, and Bilibili live streaming.
web-digital-human · znn1980 · Gitee
A browser (H5) digital human combining speech recognition, LLM dialogue through DeepSeek and others, and speech synthesis, with a Java back end.
WeClone · xming521 · GitHub
Fine-tunes an LLM on your own chat history so that a digital avatar talks like you, then connects it to chat platforms.
Xiaozhi AI (xiaozhi-esp32) · 78 (Xiaoxia) · GitHub · server
An open-source voice-chatbot firmware for ESP32 devices built on MCP, which gives smart toys, education hardware and robots voice-interactive personas; the Xiaozhi platform is run by a subsidiary of Shifang Ronghai (Tenclass).
Live-streaming digital humans
digital-avatar-project · michael7736 · GitHub
An AI live-streaming system with modular 2D and 3D avatars, speech synthesis and recognition, lip-sync, stream output and audience interaction.
digital_human_video_player · Ikaros-521 · GitHub · Gitee
A digital-human video player with an HTTP API that drives Easy-Wav2Lip, SadTalker, GeneFace++ and MuseTalk through Gradio, or plays local video.
digital_people · wukaikailive · GitHub
Fully offline digital-human live-streaming software with real-time voice interaction, custom characters, a knowledge base, lip-sync to Unreal Engine 5 MetaHumans and stream scheduling.
GPT-vup · jiran214 · GitHub
A Live2D AI streamer for Bilibili and Douyin that answers comments and Super Chats, greets viewers and thanks gift-givers; no longer maintained, and succeeded by the author’s Langup.
Luna AI (AI-Vtuber) · Ikaros-521 · GitHub · Gitee fork (hzyboy1234)
An AI virtual streamer (Live2D, Unreal Engine or xuniren) driven by many LLMs, which talks with audiences live on Bilibili, Douyin, Kuaishou, WeChat Channels, Pinduoduo, Douyu, YouTube, Twitch and TikTok, with a choice of speech synthesis and voice changers.
Video production platforms and source code
AI-Change-Background · Marthkoper · Gitee
Replaces the background of digital-human videos and images with a picture or video, in batches.
ai-digital-human-service · jsckos · Gitee
Java front and back end for a digital-human service built on Wav2Lip, with image upload and speech recognition.
AIGC digital-human SaaS · artaigc · Gitee · GitHub (gzyxds) · Gitee copy (cmsphp)
Source code for a multi-tenant digital-human SaaS aimed at short-video creators, with voice and avatar cloning, one-click presenter videos and unlimited white-label sites.
AIGCPanel · ModStart · GitHub · Gitee
A one-stop desktop app for digital-human video synthesis (MuseTalk, LatentSync, Wav2Lip, HeyGem), speech synthesis and voice cloning (CosyVoice, FishSpeech, IndexTTS, GPT-SoVITS), with local model management and more than 25 media tools; workflows and live streaming are paid extras.
aimh8_digital_human · lakysir · Gitee
A privately deployed system that queues digital-human training and short-video generation, with a WeChat mini program and a web back end for managing resellers and users.
ant-ai · caiyuanlin_13 · Gitee
A Java and Python AI application suite (image detection, video classification, speech, text-to-image and chatbots) that lists digital humans among its goals.
avatar_gpus · Rayman96 · GitHub
A Flask service that runs HeyGem inference across several GPUs, sending each video-and-audio job to the least-loaded card.
digital-person · brother-warrior-wolf · Gitee
A Vue 3 front end for a digital-human service.
digital_human · song950106 · Gitee
Front-end code for a digital-human generation platform.
Digital_Human_API · AI Horizons team · GitHub API · UI
A lecture-recording system that turns PowerPoint slides into micro-lesson videos presented by a custom digital human with a VITS voice.
Duix.Avatar (formerly HeyGem) · Silicon Intelligence · GitHub · forks: Caladog, botoai · Gitee copies: chen_jian1994, zhangjinyou
A free toolkit for making video offline: it clones a person’s appearance and voice from about 10 seconds of video, then generates talking videos from text or audio, run locally or through an API.
easegen-admin · taoofagi · GitHub
An open-source platform for making digital-human courses, combining courseware with presenter video, built on ruoyi-vue-pro, with MoFa Xingyun 3D integration for 2D and 3D video.
GenHuman · xhadmincn · GitHub
A commercial digital-human product built on vendor APIs, with web, app, WeChat mini program and admin back end, offering voice and avatar cloning and AI copywriting.
HeyGem LAN edition · Marthkoper · Gitee · fork HenGem (panshaocheng), adding batch processing and API keys
A HeyGem build for local networks that adds Chinese, English and other-language processing, subtitles, background replacement, montage and a shared public model library.
szr-bk · lyzhiwang · GitHub
An admin back end for a digital-human service, published from an admin template.
WGAI · aidy19870123 · Gitee
A free, offline AI platform bundling image and video-stream recognition (OpenCV, YOLO, OCR), speech recognition, a customer-service bot and digital-human features.
yc-digital-human · xhadmincn · GitHub · fork (cnaiart) · Gitee copy (jia611)
A digital-human system, frankly billed as a “wrapper”, that resells commercial vendors’ digital-human APIs (Silicon Intelligence, Feiying, Shanjian, Yiding) and has no models of its own.
SDKs and platform samples
alibabacloud-avatar-android-demo · Alibaba Cloud · GitHub
A sample Android app for the SDK of its virtual digital-human service.
dify-tools · Huawei Cloud · GitHub
Dify plugins, workflows and demos for its cloud products, including an interactive digital-human Q&A solution built on the MetaStudio digital-content production line.
DUIX SDKs · Silicon Intelligence · GitHub duix-sdk · Gitee duix-js · GitHub Duix-Mobile · Gitee copy of the mobile SDK (YakirZhou)
SDKs for DUIX, Silicon Intelligence’s real-time interactive digital-human platform: text- and audio-driven speech, a web rendering engine and an on-device mobile SDK for Android and iOS.
HUAWEI MetaStudio · Huawei Cloud · Gitee
The Gitee organisation for MetaStudio, its digital-human service.
MotionVerseWeb · wangCanHui · GitHub
Integration notes for embedding the Motionverse digital-human platform of DeepScience, which drives digital humans from text, voice or motion, in an iOS WebView.
NextCAS · NextHuman · GitHub server API · UE SDK
Server API SDKs (Java, Python, Go, PHP and C++) for NextHuman’s 2D and 3D digital-human open platform, and an Unreal Engine 5 SDK for its realistic, cartoon and stylised 3D digital humans.
SmartGuySdkDemo · DataBaker (标贝) · GitHub
Sample code for DataBaker’s virtual digital-human SDK.
Xingyun avatar SDK demo · MoFa · ModelScope
A runnable HTML demo of MoFa’s Xingyun real-time 3D digital-human SDK, which renders on the device instead of in the cloud, with a debugging panel, subtitle and screen settings and a full set of action commands.
ZEGOCLOUD digital-human samples · ZEGOCLOUD · GitHub examples · quick start · web demo
Sample projects for ZEGO’s digital-human PaaS: interactive chat and live broadcast scenarios, a quick start for Android, iOS, web and a Go server, and a web demo.
Voice
Amphion · The Chinese University of Hong Kong, Shenzhen, with Shanghai Artificial Intelligence Laboratory · GitHub
A toolkit for audio, music and speech generation, including text-to-speech and voice conversion, built for reproducible research.
ChatTTS · 2noise · GitHub
A generative speech model for everyday dialogue, such as an LLM assistant’s replies, in Chinese and English.
Confucius4-TTS · NetEase Youdao · GitHub
Multilingual and cross-lingual zero-shot text-to-speech that clones a voice from about three seconds of audio across 14 languages without a reference transcript; applicable to digital-human dubbing.
CosyVoice · Alibaba Group (Tongyi) · GitHub
A multilingual large voice-generation model with zero-shot voice cloning, offering inference, training and deployment.
EmotiVoice · NetEase Youdao · GitHub
A multi-voice text-to-speech engine whose emotion and style are controlled by prompts.
F5-TTS · Shanghai Jiao Tong University (X-LANCE lab) with Geely · GitHub
Fluent zero-shot text-to-speech by flow matching.
FunASR · Alibaba Group (ModelScope) · GitHub
A speech-recognition toolkit for training and inference, with streaming recognition, voice-activity detection, punctuation and speaker diarisation.
GPT-SoVITS · RVC-Boss · GitHub
Few-shot voice cloning and text-to-speech that trains a usable voice from about a minute of audio, with a web UI.
ImaginTalk · Fudan University · project page only
Generates speech, with fitting timbre and emotional prosody, from silent video of a face.
IndexTTS · Bilibili · GitHub
Controllable zero-shot text-to-speech from a single reference clip; version 2.5 covers Chinese, English, Japanese, Spanish and Arabic with fine-grained emotion control.
Linly-Dubbing · Kedreamix · GitHub · Gitee copy (homer-1943)
Multilingual AI dubbing and translation for video, with voice cloning through CosyVoice and other speech models.
MegaTTS3 · ByteDance · GitHub
Zero-shot text-to-speech with voice cloning from a short sample.
RVC · RVC-Project · GitHub
Retrieval-based voice conversion with a web UI, trained from ten minutes or less of voice data.
SenseVoice · Alibaba Group (Tongyi) · GitHub
Speech recognition for Mandarin, Cantonese, English, Japanese and Korean, with language identification, emotion recognition and audio-event detection.
so-vits-svc · svc-develop-team · GitHub
Singing-voice conversion built on SoftVC and VITS.
SoulX-Podcast · Soul App · GitHub
Generates multi-speaker podcast speech, up to about 90 minutes of continuous dialogue.
SoulX-Singer · Soul App with Geely Automobile Research Institute · GitHub
Zero-shot singing voice synthesis.
Spark-TTS · HKUST with Mobvoi, Shanghai Jiao Tong University, NetEase and others · GitHub
Text-to-speech built on the Qwen2.5 language model that generates speech tokens directly, with zero-shot voice cloning.
TTS-Core · sse-digital-man · GitHub
The speech-synthesis module of a digital-human project, switching between cloud APIs and local models behind a web UI and passing audio on for lip-sync.
Notes, guides and references
aihumanguide · willsusan · GitHub
A beginner’s guide to digital-human technology, generated with ChatGPT, as its README warns.
awesome-talking-head-generation · harlanhong · GitHub
A curated list of talking-head generation papers and code.
Awesome-Talking-Head-Synthesis · Kedreamix · GitHub
A large curated list of papers, code and resources on talking-face synthesis.
HeadNeRFonDigitalHuman · CrisHY1995 · GitHub
A Chinese write-up of HeadNeRF, a parametric head model from the University of Science and Technology of China for creating and editing high-fidelity digital humans.
I_am_a_person · yangkang2021 · GitHub · Gitee copy (812815867_admin)
Technical notes on building a real-time GPT digital human, from data preparation and avatar generation through speech recognition, LLMs, speech synthesis and driving to deployment; notes, not a project.
MetaHuman · Goulandis · GitHub
A developer’s Unreal Engine 4.27 learning project that drives a MetaHuman from an Android phone through LiveLink.
metahuman_overview · YUANZHUO-BNU · GitHub
A curated overview of digital-human technology (appearance, voice and dialogue) with representative open-source solutions for each.
QESLAT-2024 · ann-yuan · GitHub
The CCL24-Eval shared task on evaluating the translation quality of sign-language digital humans, organised by Beijing Union University with deaf-community bodies.
SelfReconIntro · jby1993 · GitHub
A Chinese introduction to SelfRecon, a CVPR 2022 oral paper that builds a high-fidelity digital avatar from a single self-rotating video.
TalkVid · The Chinese University of Hong Kong, Shenzhen, with Sun Yat-sen University and HKUST · GitHub
A large, diverse video dataset for training audio-driven talking-head models; CVPR 2026 Findings.