Build a coordinated set of candid frames around one visual point of view.
Runs up to 4 sequential still jobs — not one multi-output call. AI-generated example.
