Google just made AI video generation cheap enough that you might actually use it in production.

The Summary

The Signal

Google's Gemini Omni 1.2 Flash addresses the two things that have kept AI video generation in the demo phase: cost and clip length. The new model generates 15-second clips natively, up from 5 seconds in the previous version. That matters because 15 seconds is long enough to actually tell a micro-story, show a product in context, or create a meaningful social media asset. Five seconds was a tech demo. Fifteen seconds is a tool.

The 30% price reduction changes the economics of AI video at scale. If you're generating dozens of variants for A/B testing, personalizing video content by user segment, or automating routine video tasks, cost per clip is what kills pilot projects. Google is betting that cheaper inference drives volume, and they're probably right.

"Fifteen seconds is long enough to tell a micro-story. Five seconds was a tech demo."

The consistency improvements matter more than the press release lets on. Video generation models struggle with temporal coherence:

  • Objects morph between frames
  • Lighting shifts randomly mid-clip
  • Characters' features drift as the camera moves

Better frame-to-frame consistency means fewer regenerations to get a usable clip. That multiplies the effective cost savings well beyond 30%. You're not just paying less per generation, you're generating less to get what you need.

The Implication

This is aimed squarely at developers building video automation into products, not individual creators making art projects. The speed and cost position Gemini Omni 1.2 Flash as infrastructure for video-heavy agent workflows. Think: e-commerce product videos generated from text descriptions, personalized explainer clips for onboarding, automated social content from blog posts.

Watch for API integrations with marketing automation platforms and content management systems in the next quarter. When video generation becomes a function call instead of a creative process, a different set of use cases unlocks. The question shifts from "can we afford to make this video" to "why wouldn't we generate a video for every product, every post, every email."

Sources

Google DeepMind | Google AI Blog