Google's Gemini Omni 1.2 Flash makes 15-second AI videos for 30% less

Google released Gemini Omni 1.2 Flash, its video generation and editing model, as generally available on November 5, 2026, through the Gemini API, Google Cloud and Gemini for business. The headline changes are clips of up to 15 seconds, up from 10, and a price of $0.07 per second of 720p video, which Google says is 30% less than Gemini Omni 1.1 Flash.

Google pitches the model at “the everyday videos businesses make in volume”, naming product videos, virtual try-ons and ads, and says it follows image, video and audio references more closely than before. General availability (GA) means Google now presents it as production-ready rather than as a preview, backed by Google Cloud infrastructure and with Provisioned Throughput, Google’s option for reserving capacity.

What Gemini Omni 1.2 Flash adds over Omni 1.1

The previous version, Gemini Omni 1.1 Flash, shipped in late August 2026. Google’s two launch posts list the changes in 1.2 as follows.

• Longer clips. Videos can be generated and edited natively up to 15 seconds, up from 10 seconds with Omni 1.1, and duration can be set in one-second increments from 3 to 15 seconds.
• Lower price. $0.07 per second of 720p video, 30% less than Omni 1.1 Flash, “with similar savings at other resolutions”.
• Closer reference adherence. Users can combine character, product or style images, short motion clips of up to three seconds, and audio cues, and Google says faces, garments, packaging typography and brand colours stay consistent across camera angles.
• Less drift. Google claims stronger character, object and scene consistency from first to last frame, “with less visual drift or morphing” across motion and camera transitions, including when interpolating between a first and last keyframe or extending a clip.
• Conversational editing. A clip can be refined turn by turn, for example swapping a product or background, changing lighting or weather, or switching the aspect ratio, and this works across full 15-second clips.
• More natural speech. More fluid dialogue, clearer lip-sync and “distinct multi-speaker conditioning”, meaning separate voices can be assigned to separate characters.
• Camera control. Google says complex continuous camera moves, such as FPV (first-person-view) flythroughs, zoom-outs and orbits, come with “significantly fewer unwanted jump cuts” and steadier scale.

Spec	Gemini Omni 1.1 Flash	Gemini Omni 1.2 Flash
Maximum clip length	10 seconds	15 seconds, set from 3 to 15 in one-second steps
720p price (Google)	About $0.10 per second effective, per Google’s pricing page as quoted by e2x.ai	$0.07 per second
Resolutions	360p, 720p, 1080p, 4K	Drafts at 360p or 720p, upscale to 1080p or 4K
Motion reference clips	Not stated in sources	Up to three seconds
Watermark	SynthID	Imperceptible SynthID

How the new pricing compares

Google states the cut as a percentage against Omni 1.1 Flash. For Omni 1.1, Google’s own pricing page billed video output in tokens, $17.50 per million, with 720p consuming 5,792 tokens a second, which works out to about $0.1014 per second; the page itself printed “an effective price of approximately $0.10 per second,” according to an August audit by e2x.ai. Measured against that rounded figure, $0.07 matches the stated 30% reduction.

The same audit put Google’s Omni 1.1 rates at about $0.0338 per second for 360p, $0.1520 for 1080p and $0.3041 for 4K, so 4K cost three times as much as 720p. Google has not, in the launch text, given per-second prices for 1.2 at those other resolutions; it says only that savings are similar. Builders budgeting for 1080p or 4K output should check the live pricing page rather than extrapolate.

Google also frames the workflow itself as a cost control: draft at 360p or 720p, fix a detail with a follow-up edit instead of regenerating the whole clip, then upscale the best result to 1080p or 4K.

Where Gemini Omni 1.2 Flash is available

Google lists these access points:

• Developers: the Gemini API and Google Cloud, where the model is generally available for production use.
• Business users: Gemini for business and Google Vids, where teams can combine generated clips with avatars, music and voiceover.
• Coming soon: creating a short video from a prompt or product photo directly in the Gemini chat for business users.

One limitation is explicit: audio references are “gated behind allowlisting,” so not every account can use them yet. Outputs carry an imperceptible SynthID watermark, Google’s system for marking AI-generated media.

Several design and creative platforms are building the model into their own products, according to Google, including Artlist, FLORA, invideo, Krea, Picsart, Photoroom, QuickFrame AI and Runway. Google’s post shows Runway work made with the model and describes Runway as using it “to produce and scale high-quality content”. Third-party platforms already listed earlier versions: Picsart’s API platform, for example, shows Gemini Omni 1.1 Flash with 3 to 10 second clips and a separate preview listing for the Omni Flash family.

Google’s claims versus independent evidence

Everything in this release, the consistency gains, the speech improvements, the camera control, is described by Google. The launch posts illustrate the claims with side-by-side examples rather than numerical benchmarks. In one, Google says that given the same reference poster and a five-layer animation prompt, Omni 1.1 Flash “does not match the reference image as closely as 1.2 does”. In another, the model combines one model photo with front-and-back shots of three outfits to produce a 15-second virtual try-on that, Google says, keeps the person’s identity and each garment’s cut and drape consistent.

The other evidence on offer is customer testimony, all published inside Google’s own post, so it is not independent. A few examples:

“We ran Gemini Omni 1.2 Flash through our evals at invideo. Honestly, it’s really good. It sticks to the prompt, and the physics and sound feel real.” — Shivam Rastogi, VP of Engineering, invideo

“Version 1.2 brings 15-second videos, a top user request. With general availability and SLAs, we can rely on it in production for our sellers and enterprise customers.” — Matt Rouif, CEO and Co-founder, Photoroom

FLORA’s Andrew Taco Zhang says the model “edits exactly what you ask for, a colorway or a background, and leaves the product untouched”. Krea’s Numan Khan says it does “a great job of making only the changes we ask for while keeping everything else consistent”. QuickFrame AI’s CTO Richard Girges says 15-second generation at 1080p lets users “create a complete CTV ad in a single generation”, CTV being connected-TV advertising.

No third-party leaderboard result or independent hands-on test for 1.2 appears in the sources. Until those arrive, the improvement claims should be read as the maker’s.

What it means for builders

The 15-second ceiling matters for specific formats. Google points to complete stories, product explainers and full social ad spots in a single generation, and several partners single out the length as the practical change, for example invideo’s “15 seconds means a full scene in one go”. Fewer generations stitched together means fewer seams where characters or products can drift.

Conversational editing changes the cost model of iteration. If a background swap or lighting change can be applied to an existing clip rather than regenerating it, ad variants for different audiences and placements become cheaper to produce, which is exactly the use Google describes. Combined with low-resolution drafting, the intended loop is: draft cheap, edit in place, upscale only the winner.

Reference handling is the feature most relevant to commerce apps. Google’s examples centre on turning existing product photography or character sheets into virtual try-on clips, catalog spins and branded ads, and the Cloud post describes retailers and marketplaces turning photos they already have into product spins, lifestyle clips and try-on videos.

Limits and open questions

• Audio references are restricted. They require allowlisting, so voice-matched workflows may not be available on day one for every developer.
• Non-720p prices. Only the 720p price is stated in the launch text; other resolutions are described as having “similar savings”.
• Reseller pricing varies. For Omni 1.1, e2x.ai found the per-second 720p price moved by 44% across five providers selling the same model, and warned that a listing without a version string may not be selling the version or price it appears to. The same caution applies as 1.2 reaches third-party platforms.
• No independent quality data yet. The consistency and lip-sync claims rest on Google’s examples and partner quotes.

What to watch next

Three things will show how much of this holds up. First, independent evaluations and leaderboard placements for 1.2, which would test Google’s consistency and reference-adherence claims against rival video models. Second, the full resolution price ladder for 1.2 and how third-party platforms price it relative to Google, given how widely resellers diverged on 1.1. Third, the rollout of video creation inside the Gemini chat for business users, which Google says is coming soon, and whether audio references move beyond allowlisting.

For context on other short-form video models reaching creative platforms, see Unicorn API’s coverage of Kling 4.0 Flash on Krea.