Do you know perhaps a ready made camera movement animaton to take shots of a face from all angles?

James

Do you know perhaps a ready made camera movement animation to take shots of a face from all angle?s

AND

A face animation doing various different expressions and talking.

I want to experiment with deepfacelive.

So instead of using data from movie/clips of a person, which often not good in quality or not complete in face expression variations,
I think I can use daz3d.

crosswind

What do you want to do ? You want to swap the face in an animation rendered with DS by using DeepFaceLive ? So you firstly want to make an animation with Daz figures and expression products ?

If so, on the contrast, I don't think you can get good and natural results... Maybe it'll be better to make expression as well as talking animation with MoCap.  DeepFaceLive's models were trained with real human.. It is a question mark if it can well swap with final fusion on a 3D animation made with DS... I'm also curious about the result but I'm afraid you have to make and render such video clip which will take some time...

Besides, DeepFaceLive doesn't require facial recognition with all angels from the source video, it detects human facial characteristics in the video by using google facemsh. It's able to apply deepfake effects to a video feeds from a single camera angle, so you don't have to make it in that complicated way. For a simple test, find a video with normal quality or capture a clip of the facial movement of yourself will be much easier...

 

 

James

I want to use the character rendered video/images from DAZ to be used as the training data source.

Then I will apply the character to my face.

I experimented with deepfake training a few months ago, and finding the result was dissapointing.
Because I had to use many video sources of a person as the training data set, that had different resolution/quality, and didn't cover different angles enough, or having enough expression.
The best practice is only using one source of video, yet it has to be able to cover from different angles and face expression.
Which is almost impossible.
That's why I was thinking if I used DAZ or maybe metahuman, I could generate a perfect video/images of a person.
Although it's 3D
My guess is deepfacelive will be able to adapt it to my face, so it won't look as 3d Character.

crosswind

OK, understood. You want to train your own model. I've played with LoRA models training but have not exp. with the training in DFL... but I knew it would be a long course to have a really good trained result.

Well, I still have doubt about if it can well train models based on the source of 3D renders... with its base algorithm of real human model mesh. Good luck !