Vtuber software/hardware combinations and trade-offs?

[gomipile]

I've noticed that there's a lot of variation in animation quality, fluidity, and synchronization between different vtubers. Some have excellent everything, others not so much, even among very popular ones with agency funding, etc.

So, I'm assuming that there must be some differences in the effects of choices in hardware, software, how a model is sculpted, skinned, rigged, etc. etc. Does anyone know which sets of choices generally have different positive/negative/ymmv effects on how the final result looks and feels?

[DeadMech]

Only vaguely aware of them myself. I do know there are 2D and 3D models. The up and downside of each is probably fairly predictable.

The image recognition software that is used to track the performers body and face may, depending on how it's underlying neural network AI is trained, struggle with facets of different performers. Skin color, make up, glasses. If the image set used to train the AI didn't include enough examples of people with those features then the resulting software will have harder times detecting some people's faces.

[Brother Oni]

As I understand it, picking up the facial movements is a matter of fine tuning your lighting and camera settings so that the optimal contrast for the software's detection is obtained. I'm not aware of them going to the extent of full motion capture with key markers points on the performer's face.

That said, my wife is currently watching a Live2D tutorial video on how to map hair shake when the model's head moves and it seems to me, in addition to the camera picking up the performer's facial movements, there's also the skill of the artist making the model and the animation.
There's things from the face, hair, body, physics of these elements as the model moves (including the *cough* jiggle physics) - so it seems to me, the more effort you put into each element and ensuring that they all work together well, the better the vtuber looks.