Google releases Imagen 3 models broadly allowing users to use video content as prompts for generating context-aware images and thumbnails