“Google's Vio model performs image-to-video generation and can use a starting image as a baseline to maintain character consistency in subsequent video scenes.”