Thanks for the source information. Those projects are one of my favorites.
The screen and sweet spot position of the audience will be essential. I assume that is not something new to you, but since I write in a forum, allow me to be more detailed.
To get the camera object relation the Camera Calibrator in Cinema 4D will be of great support.
The typical idea would be to have the action in the space you want to create as animation, and the camera is positioned from that one optimal viewing angle. This needs to be rendered first. In your examples, the round corner of the screen can’t be easily (if at all, without massive effort) flattened out. This flattening process is needed, and the screen is, after all, just 2D (even bent or whatever.) The geometry is in Cinema 4D, and as your question indicates correctly, that is the better way.
Given a square pixel as a base, the rendered result needs to be projected to a flat (bent, folded as the actual screen) while holding a UV map that is even, as if the screen would be flat. The result will then be baked into the object. The result is a flat rectangular image ready to be screened.
Here is an example
I have used jpg here, not meant as a suggestion, more to keep the example small. During production, 8bit/channel has no place if the quality is vital.
You will find in this folder four files; each goes one step further in the process
Currently, the Camera Shader does not support this; otherwise, that would be a shortcut.
Let me know if you have any questions. … and please test this carefully, as I had no data from your side.
Model and mocap: Mixamo.
Click thumbnail to see full-size image