Google DeepMind released Veo 3.1's "Ingredients to Video" feature, allowing fine-grained control over video generation by specifying "ingredients" — objects, characters, scenes, styles. The user can mix and match ingredients, with the model composing them into a coherent video.