SadTalker
SadTalker copied to clipboard
Video as Source - not only for still result ?
I'm testing now the inference using Anaconda on Windows 10
I'm curious about this:
when using: --source_image <video.mp4 or picture.png
with Video file.
Does it use only the 1st frame of the video?
or is it possible to actually use the whole video frames while the face changes will be attached to it?
I did some tests but got only still image out of the video so I wasn't sure if there is a way to actually do such thing similar to how Wav2Lip work on videos and stills ?
It only uses the first frame.
For video portrait, you may need our previous work: https://github.com/vinthony/video-retalking
It only uses the first frame.
For video portrait, you may need our previous work: https://github.com/vinthony/video-retalking
Unfortunately, I didn't have much luck making "video retalking" work, I had hard time install it and make it work, I should try re-install and see if I can make it work.
I wonder if there is a chance you will get some of the retalking features to SadTalker if it's the Sad / Happy and other moods, and of course the movement itself could be powerful as extra option.
Our method can also support a little bit of emotional talking, just using a sad image or a happy image as the source frame will work, but not good enough. This will be considered as the future work.
Video-Retalking is hard to install with too many errors and also the results (via the demo collab) are flickering and not as good quality as SadTalker to be honest.
It will be GREAT to control emotions in SadTalker (maybe in Gradio changing in specific point in time) Also a BIG feature that will be great is to support not only still images but a video of course. but first allow the Gradio App to include the other inference options of course as I suggested before.
Please keep up the good work! 💙
它仅使用第一帧。
对于视频肖像,您可能需要我们之前的工作:https://github.com/vinthony/video-retalking
请问 video- retalking 项目的嘴唇抖动和人脸局部重绘导致人脸下半部变形变肿的问题有修复进展吗?