The current design is that blink copies the texture to the frame buffer before telling the app to render, hence why pages need camera permissions.

An alternative would be apps render to an rgba buffer, which we composite with camera frames browser-side (or gpu-side), but this isn't the currently implementation

We could do browser-side compositing if we add exclusive session support that draws fullscreen.  For non-exclusive with multiple canvases, I don't know an easy way to do this browser-side since we need the normal compositing flow, this would need complex hooks in compositing.