Veo 3 is Google DeepMind's most advanced video generation model, announced at Google I/O 2025. It generates high-quality cinematic videos from text and image prompts with native audio generation — including dialogue, sound effects, and ambient sound — making it the first major video model with built-in audio.