I've been learning to create 360 degree videos from 3D scenes within After Effects using the newly integrated VR Converter tools, following tutorials like
. It's working fine, not a difficult setup, however ideally I'd like to export a stereoscopic 3D VR video for YouTube with over/under mapping. I'm not using any 360 camera footage, just creating the scenes within After Effects using 3D layers.
I've discovered this tutorial which basically suggests to render the comp out normally, then shift the camera a few pixels to the side, render a second view and stitch the two results together on top of each other.
I've no doubt that this works, but I was wondering if there was a way to achieve the same effect without needing to shift and re-render a single camera setup within the scene. I.E. have two camera setups within a scene next to each other, each one feeding one half of the final over/under comp. Also, I've not experimented extensively but is there any particular art to the distance that you'd space the cameras within a scene? Just ballparking a few pixels to the side seems like it might not lead to the clearest result for the viewer at the end.
If anyone, alternatively, has a better workflow entirely for managing a 3D VR scene like this then I'd love to hear any tips.
I didn't have time to look through those tutorials, but unless something has radically changed in the way AE renders a VR view, two cameras will not do you much good: They will result in a stereo 360 view that is correct only in the front-facing view, flat in the side view (both cameras are in line) and reversed in the back-facing view (left and right eyes reversed to create the exact opposite of a correct perspective correlation for the eyes).
In order to get a proper 360 stereo VR render, you need to render at least 8 or 16 (ideally more) views for each camera with a two-camera rig that is rotatable around the center of the "head", i.e. the center point between the two cameras, render the view as you rotate around 360 degrees in either 8 or 16 steps and blend (stitch) together the center (front-facing) vertical slices of each output for each eye in a separate comp. So that will be 16 or 32 (or more) renders for one output frame.
As for the distance between the two eyes, it is safest to use real-life measurements both in the modeling and the between-cameras distance.