Gemini 3.8 Live Integration Brings Extended Reasoning to Google Veo 3

Google integrated Gemini 3.8 Live and Extended Thinking into Veo 3, allowing filmmakers to prompt and refine AI video generations through real-time voice and complex spatial reasoning as of March 2025. The update addresses long-standing continuity and physics errors in AI video generation by running chain-of-thought processing before frame rendering begins.

What's new

The update pairs Google Veo 3 with Gemini 3.8 Live and 3.8 Live Extended Thinking. When generating clips, the underlying Gemini model evaluates scene descriptions, character positioning, light physics, and camera trajectories prior to handing off the prompt to the Veo 3 diffusion engine. This extended reasoning step reduces common rendering artifacts like morphing limbs, impossible camera physics, and inconsistent background geometry.

Creators can also interact with Google Veo 3 using Gemini 3.8 Live's real-time audio interface. Instead of re-typing static text prompts, users can speak adjustments during scene layout—such as directing camera panning, adjusting lighting temperature, or modifying actor blocking—while the system updates spatial layouts dynamically before rendering final video frames.

How it fits your workflow

For directors, previsualization artists, and VFX supervisors, the integration of Gemini 3.8 Live transforms Google Veo 3 from a standard prompt box into an interactive virtual stage. Directors can talk through complex shot lists and multi-subject action sequences without wrestling with prompt engineering syntax.

Compared to standalone AI video generation models like OpenAI Sora 2 or Kling 2.1, which rely on single-pass prompt interpretation, Google Veo 3 uses Gemini's reasoning layer to plan visual continuity across multi-shot sequences. This makes Veo 3 particularly effective for storyboard-to-animatic pipelines, where spatial continuity between consecutive cuts is critical. Creators working on commercial spots or narrative shorts can refine camera blocking conversationally before committing compute credits to final high-resolution renders.

What it costs / how to try it

Gemini 3.8 Live integration in Google Veo 3 is rolling out to Google One AI Premium subscribers and enterprise API users through Google Cloud Vertex AI as of March 2025.