This is a first step towards an updated API where we pass explicit graph rather than a scene. This is because we can no longer deduct which graph to use since it will depend on a context. Will happen in several steps, so bisecting will not be such a pain.