# Soul-AILab/SoulX-LiveAct

Official inference code for SoulX-LiveAct: Towards Hour-Scale Real-Time Human Animation with Neighbor Forcing and ConvKV Memory

Repository: https://github.com/Soul-AILab/SoulX-LiveAct
Canonical: https://ross.abutalabs.com/products/soulx-liveact
Language: Python
License Family: other
Last push: 2026-06-15T09:50:41+00:00

## Health v2 (maintenance only)
Score: 54/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 87, release rhythm 35, longevity 12
- inputs: {"age_days": 174, "days_push": 79, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1176, forks 99 (observed 2026-08-28T04:03:52.376850+00:00)

## What it is
SoulX-LiveAct is the official inference code for a real-time human animation framework that generates lifelike, audio/multimodal-controlled talking-head video streams using autoregressive diffusion with Neighbor Forcing and ConvKV memory compression. It achieves 20 FPS on two H100/H200 GPUs and supports consumer GPUs like the RTX 5090 at reduced frame rates.

## Use cases
- generate real-time talking avatar video from audio
- stream hour-long digital human animations without memory growth
- run audio-driven human animation on consumer GPUs
- build a virtual podcast or talk show presenter
- deploy a real-time digital human for live interaction
- compress KV cache for long autoregressive video diffusion

## When to choose
- you need real-time, audio-driven talking-head video generation with hour-scale duration
- you have modern NVIDIA GPUs (H100/H200 or RTX 4090/5090) and want FP8/FP4 optimized inference
- you want a research-grade AR diffusion video model with constant-memory streaming

## When to avoid
- you need a simple offline video generation pipeline without real-time constraints
- you lack CUDA GPUs or sufficient VRAM for an 18B-parameter model
- you need a permissively licensed production dependency - no license is specified

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, video-processing, machine-learning, deep-learning, gpu-computing
- domain: deep-learning, artificial-intelligence, computer-vision
- platform: python
- tags: diffusion-models, human-animation, talking-head, real-time-video-generation, autoregressive-video, audio-driven-animation, digital-human, inference-optimization, video, linux, gpu

## Member repositories
- Soul-AILab/SoulX-LiveAct (main) score 54

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:52.376850+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 2026-08-30T06:26:33.852530+00:00, confidence not recorded.
  - readme: https://github.com/Soul-AILab/SoulX-LiveAct (fetched 2026-08-28T04:03:52.376850+00:00, sha 01841c989130)
- Data as of 2026-08-30T08:39:29.467469+00:00.