According to the Vertex documentation [ 30 ], the API uses an LLM-basedprompt rewriter. This means that for some tasks, the solution is likely to come from the LLM insteadof the video (e.g., Fig. 55: Sudoku). We treat the system (rewriter and video generator) as a singleblack-box entity.
given that LLM's are generalist foundation models, doesn't this call veo's "zero-shot" capabilities into question? Would like to see the prompt re-writer's outputs