Google DeepMind just made AI video cheaper, a little longer, and far easier to fix — that's the short version of Gemini Omni 1.2 Flash, which is rolling out now. The model generates clips natively up to 15 seconds, clings much harder to the reference images you give it, and costs 30% less than the version it replaces. Here's what actually changed, and what you can do with it this week.
What happened
Anish Nangia, a product manager at Google DeepMind, announced Gemini Omni 1.2 Flash. It's generally available right away through the Gemini API and Gemini for business, and it also shows up in Google Vids and Google Cloud. Nothing is "coming soon" — you can use it today.
Three changes are worth your attention.
Longer clips, priced by the second
Native video length jumps from 10 seconds in Omni 1.1 to 15 seconds. You can also dial in duration in one-second increments anywhere from 3 to 15 seconds. That's more useful than it sounds: a quick hook for a story ad and a full 15-second product explainer are different jobs, and now you don't have to fake one with the other.
The price is $0.07 per second for 720p video — 30% below Gemini Omni 1.1. Because you pay by the second rather than by the clip, a short hook costs less than a full spot, and the total is simply whatever your chosen duration adds up to at that rate. At the old price, the same clip would have set you back roughly a third more.
Editing by conversation instead of starting over
The model now supports multi-turn conversational editing. If a clip is almost right and one detail is off, you ask for the fix in a follow-up message instead of burning another full generation. Think of it as asking a photographer to reshoot one frame rather than rebooking the whole shoot.
References that actually stick
You can combine character, product or style images, short motion clips of up to three seconds, and audio cues — and 1.2 Flash holds onto those details from the first frame to the last. Faces, garments, packaging typography and brand colors stay consistent as the camera moves.
The demo Google showed is the clearest example: one model photo plus front-and-back product shots of three outfits becomes a 15-second virtual try-on sequence with 360-degree turns and match cuts, keeping the model's identity and each garment's cut, drape and back details intact. There's also stronger consistency when you interpolate between a first and last keyframe, or extend a clip into a longer sequence, with less visual drift and morphing.
What it means for you
You don't need to care about model version numbers. You need to care about what just got cheaper and what just became possible.
At home
Wedding slideshows, birthday videos, that family trip nobody ever edited — those used to require either a skill set or a subscription. A 15-second clip billed by the second changes the math. Draft at 360p to test an idea, then upscale only the take you actually like.
For work and internal comms
Short-form is where attention lives, and most teams don't have a video person. A 15-second explainer of a new process, a product update or a safety reminder becomes a draft-and-edit job instead of a production. Conversational editing matters most here: your manager says "make the logo bigger," and you fix one detail instead of rebuilding the clip.
For small businesses and e-commerce
This is the strongest use case in the release. Product photography you already own becomes motion. One model photo plus front-and-back shots of three outfits turns into a try-on sequence — exactly the asset that normally means a studio booking, a fitting model and a retoucher. If you sell clothing, accessories or packaging where the label has to stay readable, reference adherence is the feature that decides whether the output is usable.
For study and teaching
Explainer clips for a class presentation, a 15-second animation of a concept, a visual aid for a lecture. Drop in a diagram as a style reference and keep it consistent across the clip instead of accepting whatever the model improvises.
For creators
If you post short-form, 15 seconds is the native unit — a full ad spot or a complete beat in a Reel or Short. Pair a motion clip of up to three seconds with a character sheet and you get a consistent on-screen persona across shots, which is the hard part of AI video.
For side income
Cheap drafts plus a paid upscale is a workable margin. Iterate at 360p or 720p for $0.07 a second, then push only the winners to 1080p or 4K. If you freelance, that's the difference between quoting a client a reshoot and quoting them a revision.
If you're experimenting on a budget, pair paid generation with free utilities — MyKreaTool has a set of free AI tools for prompts, images and quick mockups, so you can nail the idea before you spend money on seconds.
How to try it right now
Start cheap and only spend once you know what you want.
1. Pick your entrance. Already inside Google's ecosystem? Google Vids and Gemini for business are the fastest way to see the model without writing code. Developers go straight to the Gemini API on Google Cloud, which is pay-as-you-go at $0.07 per second for 720p.
2. Write one sentence and one number. Describe the shot in plain language and set the duration between 3 and 15 seconds. Be specific about the ending — the model has 15 seconds, so decide what happens in them.
3. Attach references. Upload a character, product or style image, a motion clip of up to three seconds, and any audio cue. This is where the quality jump lives: the reference is what stops the face, the label and the brand color from drifting.
4. Draft at 360p or 720p. Don't start at the top. Cheap drafts are the entire point of this workflow.
5. Fix with a follow-up, not a reroll. When something's off, send a conversational edit. It costs less than a fresh clip and keeps what already worked.
6. Upscale the winner. Once a take earns its place, push it to 1080p or 4K.
Upsides and what changes
The obvious win is the 30% price cut plus multi-turn editing — together they cut the cost of reaching a usable clip by much more than 30%, because you stop paying for full regenerations. The second win is consistency: reference adherence across camera angles is the thing that turns AI video from a novelty into a production tool. The third is length. Fifteen seconds with one-second control is enough for a real ad, a product explainer or a full social spot, and that's the threshold where small teams can stop renting a studio.
Limitations
Be clear-eyed about it. Fifteen seconds is still fifteen seconds — this isn't a tool for a three-minute brand film, and stitching clips together takes planning. The $0.07 figure covers 720p; 1080p and 4K upscales are extra, and Google isn't advertising them as cheap. "Improved reference adherence" is a real step up, but it isn't a guarantee — packaging typography is usually the first thing to wobble, and the more references you stack, the more the model has to compromise. Conversational edits can also nudge the parts you already liked, so save your best take. And Google's own AI-generated summaries carry an "experimental" label for a reason: check the output before it goes anywhere near a client.
Conclusion
Gemini Omni 1.2 Flash isn't a sci-fi leap — it's a price cut, a duration bump and a consistency fix, which is exactly what a working tool needs. The practical effect: you can now produce a 15-second, brand-consistent clip on a pay-per-second budget at 720p, fix it by talking to it, and upscale only the take you love.
One action for today: open Google Vids, Gemini for business or the Gemini API, write a single 15-second shot you genuinely need, attach one product or character photo as a reference, generate it at 360p — then fix one thing with a follow-up message instead of regenerating. That one loop will teach you more than any demo.



Comments 0