Select Page

Gemini Omni 1.2 Flash: 15-Second Videos and Lower Costs

by | Nov 6, 2026 | AI | 0 comments

Conceptual illustration of AI video editing with consistent product imagery across floating cinematic frames

Google’s Gemini Omni 1.2 Flash update brings longer AI-generated videos, conversational editing and lower generation costs. Announced on 5 November 2026, the release is aimed at creators and businesses producing short explainers, product clips and advertising variations.

The headline is 15-second video generation, but the practical details deserve attention: API billing is token-based, some workflows retain shorter duration limits, and existing Omni 1.1 API requests are due to switch models in December.

Background: what is Gemini Omni 1.2 Flash?

Gemini Omni Flash is Google’s model for generating and editing video, including native audio. Developers can access it through the Gemini API, while Google is also rolling out related capabilities across its creative and business products.

This release follows Omni 1.1 and focuses on making short-form production more controllable. Google says it improves reference adherence, character and object consistency, camera movement and dialogue. Those are vendor claims, not findings from an independent TechKangaroos benchmark.

What changed in the new release?

Longer clips and more precise duration control

The standard generation range now extends from three to 15 seconds, selected in whole-second increments. The previous version topped out at ten seconds. That gives a creator more room for a product reveal or a short explanation without assembling separate generations.

Developers should explicitly set the duration: the documentation says the default remains ten seconds. First-and-last-frame interpolation still supports only three to ten seconds, so not every workflow gains the full new limit.

Editing through follow-up prompts

Conversational editing lets users request changes to an existing clip rather than describe the entire scene again. Google’s examples include changing backgrounds, lighting and aspect ratios while retaining the scene’s main elements.

Reference images can help anchor a character or product across shots. The API documentation allows up to six reference images. Longer clips of 11–15 seconds support one reference video of up to three seconds; shorter clips can use up to three such reference videos.

Improved consistency, according to Google

Google reports less visual drift, more natural dialogue and better control of continuous camera moves. For an online retailer, preserving a garment’s appearance or a product’s shape could be more useful than producing a spectacular but inaccurate scene. Output still needs human checking.

Gemini Omni 1.2 Flash pricing: what does it actually cost?

Google advertises a 30% price reduction and approximately US$0.07 per second for 720p video. Its pricing documentation explains that this is an effective rate, not a simple flat charge for every request.

Standard video output costs US$12 per million tokens, with 720p output consuming 5,792 tokens per second. That works out to about US$1.04 for one 15-second output, before input charges or any other applicable costs. Repeated attempts add to the bill.

For Australian teams, these are US-dollar API prices, not Australian-dollar subscription quotes. Budget for currency conversion, applicable taxes and the number of rejected takes, rather than treating the price of one generation as the cost of a finished advertisement.

The pricing page lists no free API tier for this model. Google separately says 15-second generation is coming to Google Vids at no additional cost with a Google account or Workspace subscription. Consumer-app access and API billing should not be confused.

Why it matters for creators, businesses and developers

The meaningful opportunity is faster iteration. If a useful clip needs a different setting or framing, editing it may reduce the need to start again. Whether that saves money in practice depends on how often the edited result meets the brief.

For small Australian businesses, sensible trial projects include product demonstrations, service explainers and alternative social-ad concepts. These are suggested applications, not guaranteed results. AI footage should not misrepresent a product’s capabilities or imply that a synthetic scene documents a real event.

A practical way to evaluate the update

  1. Choose one low-risk brief. Use reference assets you own or have permission to use.
  2. Define acceptance criteria. Check product shape, colours, readable text, motion and audio.
  3. Compare complete workflows. Record generation attempts, editing time and total spend.
  4. Approve before distribution. Review the final export rather than relying on an attractive preview.

Developers should also test the exact access route they intend to deploy. The announcement describes a broad product rollout, but individual API capabilities and app entitlements are not interchangeable.

Risks and limitations to understand

Google acknowledges continuing weaknesses around fast sports action, complex physics and rapid script changes. A longer clip gives errors more time to emerge, making frame-by-frame review worthwhile for commercial work.

There are also feature-specific restrictions. The API documentation says voice editing and uploaded audio references are unsupported, despite the wider announcement discussing audio-guided creative examples. Teams should use the documentation for their chosen interface rather than assume every demonstrated feature is exposed everywhere.

Video extension is separate from native generation: the documented API can append three to ten seconds per turn, up to a total of 30 seconds. It cannot prepend footage or extend the middle of a clip.

Google says generated videos include an invisible SynthID watermark. That helps identify provenance; it does not establish that depicted events are true or that the user holds all necessary rights. Obtain permission for reference material and people’s likenesses, and disclose synthetic content where appropriate.

What to watch next: the December API transition

Google’s developer documentation says requests to gemini-omni-1.1-flash will automatically route to gemini-omni-1.2-flash from 3 December 2026, with billing at the newer model’s lower rates.

Existing integrations should therefore run regression tests before that date. Compare saved prompts, reference assets, duration handling, visual consistency and downstream approval checks. Automatic routing removes a code-change requirement, but it does not guarantee identical creative output.

Conclusion

Gemini Omni 1.2 Flash makes AI video production more appealing through longer clips, follow-up editing and lower output pricing. Its value will depend on usable results, not just generation speed. Start with a controlled trial, verify the API’s specific limits and budget for review and retries.

Sources

Submit a Comment