I've set up a Stereoscopy -> Multi-View rendering to render hundreds of cameras with a Python script.
Multi-view renders the cameras much faster than looping over bpy.ops.render.render(write_still=1) because it loads the model / start Cycles only once. I'm using flat shading with 0 bounces so loading the inputs takes longer than actually rendering a camera from it.
However, this will render all cameras into RAM, and write them to disk at the end, making the computer run out of memory quickly.
If found this answer which works around the problem, avoiding the repeated model load using the Persistent Data option. That allows to use non-Multi-View at good performance. However, I would still like to know if a Multi-View solution is possible.
Is it possible to set up Blender such that the multi-view cameras will be rendered to disk instead of RAM?
