Start from a clean still. Keep copy, face guides, and graphics in the edit so they stay legible on phones. Prompt a single action and a single camera move. In a singing shot, place the intended vocal timing in the storyboard before you generate.
Check a new clip frame by frame for makeup drift and changes to hair, ear, hands, and microphone. Export a 16:9 H.264 MP4 with AAC audio, a WebP poster, and a WebVTT caption track. Make a separate mobile crop with the face kept in frame.