Google's Gemini Omni 1.2 Flash Cuts AI Video Costs 30%

Google quietly dropped a new AI video model this week, and the headline isn't the five extra seconds of footage it can generate. It's the price tag attached to them.

What Gemini Omni 1.2 Flash actually does

Gemini Omni 1.2 Flash, released on November 5, is now generally available through the Gemini API, Google Cloud's Vertex AI, and Gemini's business tools. It replaces Omni 1.1 Flash as Google's fast, cheap video-generation model — the kind built for quick turnaround rather than cinematic polish.

The headline upgrade is clip length: native output now stretches to 15 seconds, up from 10 in the previous version, with users able to dial in any duration from 3 to 15 seconds in one-second steps. Google also says the model holds a character, product, or visual style steady across the whole clip through multi-turn conversational editing — you describe a change in plain language and it revises the existing video instead of generating a fresh one from scratch.

The real story is the price cut

Google cut the cost of 720p generation to roughly $0.07 per second, about 30 percent below what Omni 1.1 charged. Billing still runs on output tokens rather than a flat per-clip rate, so the per-second figure is an effective price rather than a sticker price, but directionally it's a meaningful drop for anyone generating video at volume.

Every output still carries a SynthID watermark — an invisible digital signature embedded in the pixels that lets platforms and tools later confirm a clip was machine-generated, even if it's been re-uploaded or re-encoded. Google is also sunsetting the older model on a short runway: requests to Omni 1.1 Flash will automatically redirect to 1.2 starting December 3.

Fifteen seconds and seven cents a second doesn't sound like a lot until you realize that's roughly the length and budget of an actual Instagram Reels ad.

What this means for Indian creators and startups

This lands squarely in the market India's creator and D2C advertising economy already runs on. A huge share of the country's social video — Reels, YouTube Shorts, regional-language brand spots — is short-form by default, and small D2C brands and agencies have been priced out of traditional video production for years. A 15-second AI-generated clip at a fraction of a rupee's worth of compute changes that math, particularly for startups that can't afford a shoot but need a dozen ad variants to test.

• Indian ad-tech and marketing-automation startups building on the Gemini API get a cheaper, longer-form generation backend overnight.
• Smaller D2C sellers and solo creators can realistically test multiple video ad concepts instead of committing to one expensive shoot.
• The SynthID watermark matters locally too — India's IT Rules have been moving toward mandatory labelling of AI-generated "synthetic" media, and tools that bake in provenance markers by default are ahead of that curve rather than scrambling to retrofit it.

That labelling push is worth pausing on if you haven't followed it: the government's proposed amendments would require platforms to clearly flag content that's wholly or substantially AI-generated, similar in spirit to how deepfake disclosure rules work elsewhere. A model that watermarks everything by default is a small but real head start for Indian platforms trying to stay compliant as those rules firm up.

Where this goes next

Fifteen seconds is still nowhere near a usable ad break, let alone a scene in a film, and Google's own framing of this as a "Flash" model — fast and cheap, not flagship — suggests it's not meant to replace higher-end generation tools. But the pattern is familiar from every other corner of generative AI: the frontier model gets the demos, and the cheap, fast model is what actually gets embedded into products people use every day. For India's crowded field of short-video apps and ad-tech tools, that's the version that matters.