How to Create a Personal AI Avatar in Doubao
A Juejin hands-on guide shows how Doubao mobile users can create a personal AI avatar from facial and voice samples, then use it in generated videos. It documents the creation workflow, optional selfie uploads, lighting and recording tips, plus restrictions including mobile-only access, liveness verification and use of one’s own face. The author says the avatar reproduced their appearance and speaking manner, but this realism assessment is not independently verified.
The guide provides a concrete workflow for creating and using a personalized video avatar while highlighting consent, privacy and platform safeguards.
前言
上篇文章笔者分享了视频生成提示词的编写技巧。文章发出后,相关交流群里讨论得热火朝天。有读者半开玩笑地说:“现在借助提示词生成的视频效果都这么好了,能不能让我也上去演一把?”
你别说,答案还真就是:能!借助豆包就可以。
笔者第一次用这个功能生成完视频时,盯着屏幕看了很久:画面里的人是我,发型也是我的,连我说话时习惯性抿一下嘴的小动作都在。可那天现实中,我还穿着睡衣、头发乱着,根本没坐到镜头前。
这究竟是怎么做到的?下面大家就跟笔者一起来看一下吧~
一、真人分身并不是真人换脸
真人分身就是基于你的面部和声音特征,生成一个能替你出镜的专属AI分身。以豆包为例,你只需按提示录制一段短视频:向左转头、向右转头、对着镜头朗读几个数字,并自然微笑。系统会抓取你的面部关键点与骨骼比例、微表情、肌肉动态,以及声音的音色和韵律,再由豆包背后的Seedance大模型,通过实时生物特征捕捉、3D面部建模、声纹特征迁移等关键技术,建立专属的分身模型。