Humanlike AI
全部模型
HumanLike 标志

HumanLike Avatar Forever

AI 数字人模型 · 预览版

HumanLike

One photo and a voice track in, a talking person with hands and gestures out.

HumanLike Avatar Forever待机 · 示例即将上线

驱动部位

招聘适配度

70

每小时

$60.00

每分钟

$1.00

30 分钟面试

$30.00

简介

Avatar Forever is the rendered counterpart to Homo 1. It runs the open Avatar-Forever checkpoint (LetsThink, built on LTX-2.3 22B) on HumanLike’s GPUs and animates the whole upper body from the voice: head, mouth, shoulders and hands. It is a video model, not a live one. Measured on one H100 in October 2026, 4 to 7 seconds of 1024x576 video take 70 to 100 seconds, about 15 times the spoken length, so it makes recorded answers, intros and video messages, and it cannot hold a live call.

  • Rendered video
  • Hands and gestures
  • Lip-sync from any TTS
  • Custom faces

驱动范围

面部
完整
口型同步
完整
上半身
完整
双手
部分
全身
无
提示词生成内容
部分

Face, lips, head and upper torso are driven by the voice and the prompt; hands gesture when the prompt asks for it. Each segment is a separate render from the same first frame, so exposure and small props can shift between segments.

来源:HumanLike renders on Modal, 25 to 29 October 2026

优势

  • Hands and upper body move
  • Runs on HumanLike GPUs
  • Open weights

注意事项

  • Not live
  • About $1 per spoken minute
  • Exposure can shift between segments

最适合 Recorded candidate messages · Job intro videos

数据

公开基准

SyncNet confidence
7.7自测

Global SyncNet confidence on a 26 s render with zero measured drift. The released audio mapping scored 3.3 on the same segment.

对比