- Blog
- Gemini Omni 1.2 Flash: 15 Seconds for $1.05, and an End to the Re-Roll
Gemini Omni 1.2 Flash: 15 Seconds for $1.05, and an End to the Re-Roll

Gemini Omni 1.2 Flash is now generally available on the Gemini API and Gemini for business, and the headline numbers are easy to repeat: 15-second clips, and $0.07 per second for 720p, a 30% cut from Omni 1.1.
Both are real improvements in Gemini Omni 1.2 Flash. Neither is the reason to care.
The Gemini Omni 1.2 Flash reason to care is that editing a clip no longer means generating it again. Gemini Omni 1.2 Flash supports multi-turn conversational editing on full 15-second clips — swap the product, change the lighting, switch the aspect ratio, all as follow-up turns. That converts video generation from a repeated gamble into an iterated draft, and it changes the cost of being wrong far more than the price per second does.
What Shipped in Gemini Omni 1.2 Flash
Native 15-second generation in Gemini Omni 1.2 Flash, up from 10 seconds in Omni 1.1, with duration settable in one-second increments from 3 to 15.
Multi-turn conversational editing across the full clip length of Gemini Omni 1.2 Flash.
Stronger reference following in Gemini Omni 1.2 Flash — character, product or style images, short motion clips of up to three seconds, and audio cues, all combined. Google specifically claims faces, garments, packaging typography and brand colours stay consistent across changing camera angles.
Cleaner multi-speaker dialogue — distinct voice conditioning, so each sampled voice is routed to the right character, with clearer lip-sync.
Better camera control — continuous moves like FPV flythroughs, zoom-outs and orbits with fewer unwanted jump cuts.
Enterprise readiness — the first time Gemini Omni 1.2 Flash is generally available, on Google Cloud infrastructure, with Provisioned Throughput and an imperceptible SynthID watermark on output.
The Price Is Good. The Cost Model Is Better
Read Google's own description of the workflow, because it contains the more important change:
draft fast at 360p or 720p, fix a detail with a follow-up edit instead of re-generating the entire clip, and then upscale your best creations to 1080p or 4K.
That is a three-stage pipeline, and each stage attacks a different waste.
Drafting cheap. 360p drafts exist so you can burn attempts on composition rather than pixels.
Editing instead of regenerating. This is the big one. In the old model, one wrong detail meant one more full-price clip, and a small chance the re-roll broke something that was previously fine. With Gemini Omni 1.2 Flash, you fix the detail and keep everything else.
Upscaling only the winners. With Gemini Omni 1.2 Flash you pay for high resolution once, at the end, on the clip that already works.
The Gemini Omni 1.2 Flash price per second is the least interesting number in that list, because it is the only one that is linear. The other two reduce how much you generate at all.
Doing the Actual Math
The per-second price converts cleanly.
In Gemini Omni 1.2 Flash, a full 15-second clip at 720p is $1.05. A minute of finished video is $4.20.
Compare that with the video pricing we covered earlier: Seedance 2.5 works out to roughly $0.10 per second at list rates for a 5-second 480p clip. On sticker price, Gemini Omni 1.2 Flash at $0.07 is meaningfully cheaper at a higher resolution, which is what a 30% cut and a 720p baseline will do.
Now add the re-roll factor. Sketch the old way of working: generate ten clips at 15 seconds, reject four for a bad hand, a morphing logo or a wardrobe change. That is $14.70 spent, with nine seconds of usable footage per rejection to show for it.
With Gemini Omni 1.2 Flash, the same four rejections might be four edits — changing a hand, correcting a logo, restyling a garment — keeping the ten good clips you already paid for. The saving is not 30%. It is whatever fraction of your generations used to be corrections.
For anyone producing at volume — a 60-episode vertical series, say — this compounds. Sixty 15-second episodes is 900 seconds, about $63 of generation at 720p in a single pass. The variable that actually determines the budget is not that $63. It is how many times you have to go back.

Multi-Turn Editing Is the Real Upgrade
Google's own demonstrations make the shift legible.
One example takes a single generated clip and swaps the background turn by turn — striped pavilion, oak forest, stable with a white horse, market fair, stone castle — while holding the characters' performances, armour reflections and camera framing.
Another starts with a rippling octopus and replaces it in the next turn with a dancer whose gown mirrors the exact fluid motion.
That is a different authoring model, and it is what Gemini Omni 1.2 Flash actually sells. You are no longer sampling outputs and selecting. You are directing one artefact over several turns, and delivering ad variants as Gemini Omni 1.2 Flash edits rather than fresh generations at full price.
It is also the video version of a shift Gemini Omni 1.2 Flash makes explicit, already visible in still image work, where local editing quality — not raw generation quality — has become the differentiator.
References Are Now the Production Method
The most commercially direct claim in the launch is reference adherence, and the example is a virtual try-on.
Combine one model photo with front-and-back product shots of three outfits, and Gemini Omni 1.2 Flash generates a 15-second try-on sequence across 360-degree turns and match cuts, holding both the model's identity and each garment's cut, drape and back details.
If Gemini Omni 1.2 Flash holds up outside a demo, the implications are concrete: existing product photography becomes motion content without a reshoot. That is the same economic promise the advertising industry has been chasing all year, delivered through a model instead of a governance platform.
The details to test are unglamorous: does packaging typography survive a camera move, and do brand colours stay exact after a reframe.

Multi-Speaker Dialogue Is the Quiet Win
Clean lip-sync is table stakes now. Distinct multi-speaker conditioning is not, and Gemini Omni 1.2 Flash adds it.
Google describes Gemini Omni 1.2 Flash routing each sampled voice to the correct character given a scene image plus character, mouth and one-second voice references, with expressive timing across a 15-second scene. Applied in Flow, Gemini for business, and Google Vids.
For explainers, corporate video and anything with two people talking, this is the capability that was previously the reason to avoid AI video altogether.
Read the Launch With Care
Four honest qualifications.
Every example is Google's. All Gemini Omni 1.2 Flash demonstrations, including the Pantheon FPV flight with zero jump cuts, the correctly spelled motion-tracked callouts, the try-on sequence — these are curated demonstrations chosen by the vendor, not a random sample.
$0.07 is 720p. 1080p and 4K come through upscaling, and that step has its own cost and quality characteristics. The headline rate is the middle of the pipeline, not the end.
15 seconds is native in Gemini Omni 1.2 Flash, not unlimited. Extending beyond it is a separate operation, and extension is exactly where drift historically creeps back in.
There are no independent benchmarks of Gemini Omni 1.2 Flash yet. Claims about reference adherence and reduced morphing are the kind that need a third party with a fixed prompt suite, and this launch does not have one.
What to Actually Test
Four Gemini Omni 1.2 Flash checks worth your own time.
One reference-heavy product shot. Does the logo survive a 360-degree turn?
One two-person dialogue. Is each voice attached to the right character across the cut?
One edit chain. Take a good clip and make five successive edits. Watch whether detail degrades with each Gemini Omni 1.2 Flash turn — multi-turn edits can compound error just as multi-turn prompts can.
One 15-second continuous camera move. Count the jump cuts yourself rather than trusting the demo.
A Prompt to Try
Reference adherence is best tested with a subject that has a signature detail nobody can fake — typography, a specific graphic, a particular material.
Copy & Paste this into Fanch AI (Model: gpt_image_2):
A clean studio product photograph of a rectangular matte-black aluminium beverage can standing on a pale sand plinth against a seamless warm grey backdrop: subtle brushed metal texture on the body, a plain unmarked label band in deep teal wrapping the middle, condensation-free clean surface, soft directional studio light from the left with a gentle contact shadow. Precise edge definition, fine detail in the metal grain, 16:9 framing, 8K detail, no people, no brand logos, no text overlay.
Feed that in as a Gemini Omni 1.2 Flash reference and see whether the teal band stays the same teal after a camera orbit. That is the Gemini Omni 1.2 Flash test that matters.
Why Fanch AI Fits
Gemini Omni 1.2 Flash makes the reference frame the production method: character sheets, product photography, style images. Which means the still you start from now decides what your video can hold together, and it is the cheapest thing in the whole pipeline to get right.
Fanch AI gives you an image model you direct with precise prompts, so you can build clean, consistent reference frames and character sheets before you spend anything on seconds of video.
