In engineering practice, a hybrid approach of "multi-clip splicing + voiceover" is often more pragmatic and stable than pursuing end-to-end pure video generation.