# anliyuan/Ultralight-Digital-Human

一个超轻量级、可以在移动端实时运行的数字人模型

Repository: https://github.com/anliyuan/Ultralight-Digital-Human
Canonical: https://ross.abutalabs.com/products/ultralight-digital-human
Language: Python
License Family: other
Last push: 2026-07-22T12:29:48+00:00

## Health v2 (maintenance only)
Score: 64/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 93, release rhythm 35, longevity 49
- inputs: {"age_days": 692, "days_push": 42, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2627, forks 387 (observed 2026-08-28T04:07:05.385010+00:00)

## What it is
An ultralight talking-head (digital human) model that animates a person's face from audio input and runs in real time on mobile devices. It uses lightweight audio encoders (wenet/HuBERT) with a UNet-based video generation pipeline, with training and streaming inference scripts in Python.

## Use cases
- create a talking avatar from a short training video
- run a real-time digital human on a phone
- generate lip-synced video from audio
- build a lightweight virtual presenter or chatbot avatar
- stream a digital human with low latency on mobile
- train a custom talking-head model from my own footage

## When to choose
- you need a very lightweight, real-time talking-head model on mobile or edge devices
- you want an open-source alternative to commercial digital human APIs
- you have clean audio/video training footage and want a custom avatar
- you need streaming inference with low per-frame latency

## When to avoid
- you need high-fidelity, full-body or expressive avatars rather than a single talking face
- your source audio is noisy or has echo, which degrades results significantly
- you need a production-ready, well-licensed package (no explicit license file)
- you want a maintained C++ deployment path out of the box (see the successor FeatherTalk instead)

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, speech-recognition, video-processing, llm-inference
- domain: deep-learning, computer-vision, speech-processing, mobile-development, artificial-intelligence
- platform: python, cross-platform
- tags: digital-human, talking-head, audio-driven-video, real-time-inference, onnx, mobile-deployment, wav2lip-alternative, streaming-inference, android, ios, gpu

## Member repositories
- anliyuan/Ultralight-Digital-Human (main) score 64

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:05.385010+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:19:43.953749+00:00, confidence not recorded.
  - readme: https://github.com/anliyuan/Ultralight-Digital-Human (fetched 2026-08-28T04:07:05.385010+00:00, sha f421ea59c760)
- Data as of 2026-08-30T08:39:29.467469+00:00.