AI Street Interview Vox Pop Ad Workflow

Prompting rules for AI street interview vox pop ad: > one face per clip. the interviewer only exists as a hand and a mic > every product already in their hands in the frame. "pull it out of the bag" = cloned products > under 10 words per 8s take, or the model starts looping its lines Workflow: - nano banana 2.1 for the start frames - omni inside google flow for every talking shot - trim omni takes into 38 seconds, killing every dead pause - time captions word by word against raw audio so text and lips never drift - catch object disappearing mid-sentence and bury under punch-in zoom - spot interviewer's voice bleeding into last word, cut it, drop questions back in from separate take - balance 6 different AI voices so the whole thing sounds like one mic on one street