论文地址：https://www.microsoft.com/en-us/research/project/vasa-1/

DiT约束条件

引入了5个约束条件，包括主要注视方向、头部到摄像头的距离、情绪偏移量、为了增加帧间平滑性，还增加了前K帧的语音特征和生成的运动特征，具体如下：

condition signals:

1. main gaze direcction $g=(\theta ,\varphi )$ [70], focused direction of the generated talking face.
2. head-to-camera distance $d$ [16]: normalized scaler controling the distance between the face and camera, affecting the face scale in the
   generated video.
3. emotion offset $e$ [41]: modulates the depicted emotion on the talking face.
4. last $K$ frames of the audio feature, $A^{pre}$
5. last $K$ frames of the generated motions, $X^{pre}$

MegaPortraits