# Soul-AILab/SoulX-FlashTalk

SoulX-FlashTalk is the first 14B model to achieve sub-second start-up latency (0.87s) while maintaining a real-time throughput of 32 FPS on an 8xH800 node.

Repository: https://github.com/Soul-AILab/SoulX-FlashTalk
Canonical: https://ross.abutalabs.com/products/soulx-flashtalk
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-07-30T09:14:06+00:00

## Health v2 (maintenance only)
Score: 58/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 95, release rhythm 35, longevity 17
- inputs: {"age_days": 251, "days_push": 34, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1479, forks 138 (observed 2026-08-28T04:04:50.688545+00:00)

## What it is
SoulX-FlashTalk is a 14B audio-driven talking avatar model that streams infinite real-time video from a reference image and audio, achieving 0.87s startup latency and 32 FPS on an 8xH800 node. It ships inference code and Hugging Face model weights under Apache-2.0.

## Use cases
- generate a talking avatar video from a photo and audio
- real-time streaming digital human for live streaming
- build a video podcast presenter from audio
- low-latency audio-driven talking head generation
- infinite-length avatar video generation
- virtual host for livestreams

## When to choose
- you need real-time, infinite-length audio-driven avatar video with sub-second startup
- you have multi-GPU (8xH800-class) infrastructure
- you want an Apache-2.0 model with released weights and inference code

## When to avoid
- you only have a single consumer GPU - use SoulX-FlashHead instead
- you need training/fine-tuning code, which is not released
- you need a hosted online demo, which is not yet available

## Facets
- artifact type: library
- maturity: active
- function: video-processing, audio-processing, machine-learning, llm-inference, gpu-computing
- domain: artificial-intelligence, computer-vision, deep-learning
- platform: python
- tags: talking-head, audio-driven-avatar, streaming-inference, video-generation, digital-human, real-time, model-weights, inference, video, audio, linux, gpu, docker

## Member repositories
- Soul-AILab/SoulX-FlashTalk (main) score 58

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:50.688545+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 2026-08-30T04:34:17.856201+00:00, confidence not recorded.
  - readme: https://github.com/Soul-AILab/SoulX-FlashTalk (fetched 2026-08-28T04:04:50.688545+00:00, sha 7b25d621e55f)
- Data as of 2026-08-30T08:39:29.467469+00:00.