Models

Gemini Omni 1.2 Flash gives you 15-second clips and a 30% lower bill

Google made Gemini Omni 1.2 Flash generally available on November 5, and two changes matter more than the rest: clips now run 15 seconds, and the per-second price fell 30%. Both land on the same line of a production budget.

Daniel HarrisDaniel Harris
Heat: 1,420
Abstract film strip frames fading from violet to amber

Google made Gemini Omni 1.2 Flash generally available on November 5, and two changes matter more than the rest: clips now run 15 seconds, and the per-second price fell 30%. Both land on the same line of a production budget.

What 15 seconds buys you in Gemini Omni 1.2 Flash

Ten seconds was the old ceiling, and ten seconds is an awkward length. It's long enough to look like an ad and too short to work as one, so editors stitched two clips together and lived with the seam.

Fifteen seconds fits a complete social spot: product on screen, one line of copy, a logo at the end. Duration can be set in one-second steps from 3 to 15, which helps when a platform wants exactly 6 seconds and you'd rather not trim in an editor afterwards.

The practical win is fewer regenerations. A cut that fits in one pass costs one generation instead of two, and the second generation was never the expensive part anyway. The review time was.

That's the invisible cost.

Gemini API pricing drops 30 percent

Pricing is the other half of the story, and the new rate is $0.07 per second for 720p video, down 30% from Gemini Omni 1.1. A full 15-second clip lands near $1.05 at that rate.

Put ten concepts through review and you've spent about $10.50 before anyone approves a frame. That's the number to bring to a budget meeting, because it makes iteration cheap enough to do on purpose instead of in a panic.

Google calls the price cost-effective. That's a phrase from a press release. The arithmetic is what matters here, and the arithmetic says a small studio can now afford to throw away nine drafts to find the tenth.

The Google video model workflow Google recommends

Workflow matters as much as raw clip length, and Google's suggested pipeline is draft cheap, edit cheap, upscale late. Start at 360p or 720p, fix details with a follow-up edit instead of a fresh generation, then upscale the takes you like to 1080p or 4K.

That order matters more than it sounds. Regenerating a 15-second clip to change a jacket colour throws away the whole take. A conversational edit keeps the camera move and swaps the wardrobe.

Multi-turn editing works on the full clip: replace the product, change the background, move the light from noon to dusk, or switch aspect ratio. Each of those used to mean starting from an empty prompt.

Reference adherence in Gemini Omni 1.2

Consistency decides whether a clip is usable, and the model takes reference images for characters, products and styles, plus motion clips up to three seconds and audio cues. Google says faces, garments, packaging typography and brand colours stay consistent across camera angles.

Brand colours holding is the claim agencies will test first, because a shifted brand red is the kind of error a client spots in two seconds. Google published comparison clips against Omni 1.1 Flash that show cleaner results. They're vendor demos, not independent tests, and no outside lab has scored them.

First-to-last frame consistency also improved, with less morphing across transitions, and the model can interpolate between keyframes or extend an existing clip. Speech got clearer lip-sync and better multi-speaker timing, which matters for anything with dialogue in it.

Where Gemini Omni 1.2 Flash runs

Availability decides who can try this without switching tools, and the model is generally available on the Gemini API and in Gemini for business. It also shows up in Google Vids, Flow and Google Cloud, so a team already inside Workspace can try it without adding a vendor.

That distribution is the quiet advantage. A marketing team in Chicago that lives in Google Vids doesn't need to evaluate a startup's dashboard. It gets the model inside a tool someone already opens on Monday morning. Distribution beats benchmarks here.

What the Google video model still can't do

Limits are easier to list than strengths, and audio is better here without being studio quality. Don't plan a campaign around generated voiceover without a human pass at the end.

Physics stays approximate, so a pour, a splash or a fabric fold can still read as wrong on a close look. Feature-length work is out of scope, and nothing here generates licensed music.

So what should you expect from a 15-second clip? A usable first draft, not a finished frame. The model shortens the distance between an idea and something watchable. It doesn't remove the people who decide whether the result is any good.

What a small studio should test in Gemini Omni 1.2 Flash

A cheap test beats a long evaluation, so pick one product, one character and one 15-second script. Generate three takes at 720p, edit the best one twice instead of regenerating, then upscale it.

If the brand colours survive the upscale, you've learned the most useful thing about the model for your own workflow. If they don't, you've spent about three dollars finding out.

Cheap test, clear answer.

Why Gemini Omni 1.2 Flash matters beyond the demo

The interesting shift isn't quality, because quality claims from a vendor demo are worth what you paid for them. The shift is that a 15-second native clip removes a stitching step that used to cost an editor an afternoon.

Teams that budget video by the asset will feel this first. Teams that budget by the campaign will notice the lower line item next quarter, when someone adds up how many drafts it took to get one approved cut.

Google hasn't said whether Omni 1.1 stays available at the old price. Anyone with a pipeline pinned to that model should check the deprecation notes before the next billing cycle, because a 30% swing in either direction changes what you can promise a client.

Share This Story

Sources

Related AI News

A terminal window with a rising reward curve and a 27B badge beside it
Models

RadixArk Fine-Tuned a 27B Model on 1,200 Tasks and It Beat the Original Everywhere

RadixArk took an existing open model, spent 18 days and about 1,200 hand-built tasks on it, and changed nothing about its shape. The gains are real, and the absolute scores are still low.

Heat: 1,330
A laptop outline with three stacked layer sheets passing through a loop arrow inside it
Models

Inkling 2 Mini Fits on a Laptop by Running the Same Layers Three Times

Thinking Machines Lab built a model that thinks longer by running the same block of layers three times. The trick buys reasoning without extra memory, and the bill comes due somewhere else.

Heat: 1,710
A long teal audio waveform with a single orange accent bar on a dark field
Models

A 0.5B Speech Model Just Matched a 30B One on Real Phone Calls

Reson8 shipped a half-billion-parameter transcription model that gets within a tenth of a point of a 30B rival on noisy phone calls. The interesting part is what that does to a per-minute bill.

Heat: 560
Three neon rings orbiting a small glowing core on a deep blue field
Models

Gemini 4 Neon Lands in the App, and the Model Picker Loses Two Names

Google's fourth Gemini model in a month slots into the middle of the lineup, priced to be used every day. The app now asks subscribers to choose by effort, and two familiar names are gone.

Heat: 1,500