
HumanLike Avatar Forever
AI 数字人模型 · 预览版
HumanLikeOne photo and a voice track in, a talking person with hands and gestures out.
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.