Google made Gemini Omni 1.2 Flash generally available on November 5, 2026, across the Gemini API and Gemini for business. The release extends native video generation and editing to 15 seconds, adds multi-turn conversational editing and cuts the published 720p price by 30% to $0.07 per second.
For production teams, the useful change is the combination: longer clips, more ways to preserve references and a workflow built around revising an existing clip before upscaling the selected result. Google describes these capabilities in its launch announcement.
The access distinction matters, though. General availability covers the API and business offerings. Access in the consumer Gemini app is still rolling out to Google AI Ultra subscribers.
Where Developers Can Use Gemini Omni 1.2 Flash
Google explicitly positions the Gemini API and Google Cloud availability for production use. Developers can build video generation into applications, while business users can access the model through Google’s business offerings.
The Gemini Omni 1.2 Flash model card lists a broader distribution footprint:
- Gemini App and Gemini Enterprise App
- Google AI Studio and Gemini API
- Google Cloud / Vertex AI
- Google Flow and YouTube
- Google Workspace
That list identifies distribution channels. It should not be read as a guarantee that every interface has identical controls, limits or rollout timing. In particular, the ongoing Ultra rollout means a developer’s API access and a colleague’s Gemini app access may arrive on different schedules.
For a production integration, I’d make the API or relevant business product the planning baseline. Waiting for the consumer app adds an availability dependency that Google’s general-availability announcement already removes for those other routes.
General availability also should not be treated as unlimited capacity. The supplied announcement does not give numerical throughput, concurrency or latency guarantees. Teams still need to check the limits and commercial terms attached to their chosen service before scheduling bulk production.
What the 30% Price Cut Actually Buys
Google’s published rate is $0.07 for each second of 720p video, down 30% from Omni 1.1. That implies a previous equivalent rate of $0.10 per second, calculated as $0.07 divided by 0.70.
Using those rates, the base generation arithmetic looks like this:
| Measure | Omni 1.1 | Omni 1.2 Flash |
|---|---|---|
| Maximum native clip duration | 10 seconds | 15 seconds |
| 720p price per second | $0.10, implied by the stated reduction | $0.07 |
| One 10-second clip | $1.00 | $0.70 |
| One maximum-length clip | $1.00 for 10 seconds | $1.05 for 15 seconds |
These are calculations from Google’s announced pricing and duration limits, not independently measured production costs.
The useful comparison has two parts. At the same 10-second duration, the saving is straightforward: 100 generations fall from $100 to $70. But a team that immediately adopts the new maximum length spends $1.05 per clip, compared with $1.00 for the old maximum.
In other words, 50% more running time costs 5% more per maximum-length generation. The unit-price improvement is substantial, but choosing longer clips can absorb the cash saving.
A second budget variable is the number of attempts needed to get an approved result. Twenty 15-second candidates would total $21 at the published 720p rate, before any separately priced operations. That is a more useful planning figure than treating $1.05 as the cost of a finished advertisement.
The announcement does not provide separate rates for 360p drafting, conversational edits or upscaling. Those operations therefore cannot be folded into a reliable all-in price from this announcement alone.
Draft, Edit and Upscale Only the Selected Clips
Google recommends a workflow that starts at 360p or 720p, uses follow-up edits to fix details and then upscales selected results to 1080p or 4K.
That sequence separates creative exploration from final delivery:
- Draft the scene: establish the composition, action and timing.
- Refine the candidate: change a background, product, lighting condition or aspect ratio through conversation.
- Approve the content: check references, continuity and dialogue.
- Upscale the selected result: create the higher-resolution version needed for delivery.
The practical appeal is avoiding high-resolution finishing work on every rejected candidate. However, the exact savings depend on the operation-specific pricing and how often the model produces an acceptable result.
Google says multi-turn editing works across full 15-second clips. Its examples include successive background changes while keeping character performances and camera framing intact. Other supported edits described in the announcement include product swaps, weather changes and aspect-ratio changes. See Google’s editing and workflow examples.
For ad production, that could turn one approved scene into the starting point for several variants. A team might retain a character’s action while changing the setting for a different audience or placement.
My view is that preserving an approved performance could be more valuable than the extra five seconds of duration. When timing and framing already work, rebuilding the whole scene introduces more opportunities for something else to change. The value of conversational editing depends on how reliably it preserves those approved elements.
Longer Clips Give References More Work to Do
Omni 1.2 Flash supports native generation and editing from 3 to 15 seconds in one-second increments. The previous maximum was 10 seconds.
That lets a production team request a specific duration within the supported range, instead of always generating the longest clip and trimming it later. A 6-second placement and a 15-second product explainer can use different generation lengths from the outset.
Google also says the model follows combined references more closely. Its supported guidance includes:
- Images: character references, product photography and visual styles.
- Motion clips: short video references up to three seconds long.
- Audio cues: guidance for the generated audiovisual scene.
One published example combines a model photo with front-and-back product shots of three outfits to produce a 15-second virtual try-on sequence. Google says the result maintains the model’s identity and garment details through turns and match cuts.
The production implication is specific: reference adherence could reduce the need to choose between a good performance and an accurate product depiction. A clip with convincing motion still fails a catalog brief if the garment changes shape or the packaging text drifts.
I’d treat identity, logos, typography and garment details as acceptance checks for every candidate. Those are precisely the details Google highlights, and small errors there can make an otherwise polished clip unusable.
Quality Gains Still Need Shot-Specific Checks
Frequently Asked Questions
3 questions
1How Much Does a 15-Second 720p Clip Cost?
At Google’s published $0.07-per-second rate, a 15-second generation costs $1.05. That calculation excludes any separately priced editing, upscaling or other operations.
