Models
Gemini Omni 1.2 Flash Stretches Clips to 15 Seconds and Cuts Price 30%—But Skips the Numbers That Matter
Google lists 15-second clips and $0.07 per second at 720p for Gemini Omni 1.2 Flash, but publishes no speed, quality or full-workflow cost data. Here is what to test.
Google DeepMind says Gemini Omni 1.2 Flash raises the maximum generated clip from 10 seconds to 15 and costs $0.07 per second for 720p video, a 30% price reduction from Gemini Omni 1.1. 1 That is a clear spec change. The announcement does not report measured generation speed or full-workflow costs, or establish how reliably the model performs on a particular task. 1
Google announced the model on November 5, 2026, and says it is generally available across the Gemini API and Gemini for business. 1 The announcement specifies clip length, a 720p price, and an iterative editing workflow. It does not provide measured speed, benchmark results, or full-workflow cost figures. 1
What actually changed from Gemini Omni 1.1 to 1.2 Flash
Duration is the clearest upgrade. Google says Gemini Omni 1.2 Flash can generate and edit videos natively up to 15 seconds, with durations from 3 to 15 seconds available in one-second increments. Gemini Omni 1.1 had a 10-second maximum. 1 That extra headroom may help when an output fits in one clip; it does not establish whether longer generations run faster or fail less often.
| Capability | Gemini Omni 1.1 | Gemini Omni 1.2 Flash | What the announcement provides |
|---|---|---|---|
| Maximum stated clip duration | Up to 10 seconds 1 | Up to 15 seconds 1 | A longer stated maximum 1 |
| 720p price | Not stated 1 | $0.07/second 1 | Google says the price is 30% lower than 1.1 1 |
| Measured generation speed | Not stated 1 | Not stated 1 | No speed figures or latency comparison 1 |
Price and duration are shortlist inputs, not a verdict. Google identifies the Gemini API and Gemini for business as availability channels; that does not establish whether the model meets a team's throughput, quality, or workflow requirements. 1
Gemini Flash pricing: what the $0.07 per second covers
The key price figure in the announcement is $0.07 per second for 720p video, which Google says is 30% lower than Gemini Omni 1.1. 1 The announcement gives no prices for other resolutions. The 720p rate is a starting point for comparing one output configuration, not a full-workflow cost estimate.
Google describes a workflow that drafts at 360p or 720p, uses follow-up edits to fix details, and optionally upscales selected results to 1080p or 4K. 1 The announcement does not specify whether the listed 720p rate applies to each generation and edit pass, or give prices for 360p, 1080p, 4K, or upscaling. 1
Analysis: Budget for the workflow your team would actually run—not just clip duration multiplied by the listed rate. Track draft attempts, edits, rejected outputs, selected outputs, and any upscale pass. That can help quantify costs not specified in the announcement; it is not a claim about how Google bills those steps. For a broader view of why a lower list price is not the same as lower total cost, see computational arbitrage as an allocation problem, not a savings claim.
Is the Gemini video model actually faster?
The announcement provides no measured generation-speed figures, latency breakdowns by clip duration, resolution, or task complexity, benchmark scores, or independent test results. 1
A longer maximum duration does not by itself establish that a production workflow finishes sooner. Analysis: Fifteen seconds might let some teams avoid stitching together separate segments, but whether that saves time depends on generation latency, retries, and editing effort—none of which the announcement measures. 1
During evaluation, track two separate measures: time to first usable result and total time to an accepted clip. The first reflects request speed; the second includes edits and retries. The launch post does not provide either measure. 1
What's claimed about video quality—and what isn't proven
Google says generation can be guided by character, product, or style reference images, motion clips up to three seconds, and audio cues. 1 It also describes multi-turn conversational editing—for example, swapping a product or background, changing lighting or weather, or switching aspect ratio—across clips up to 15 seconds. 1
The post claims improvements in character, object, and scene consistency, as well as speech, lip-sync, and camera control. It illustrates these claims with a 15-second virtual try-on, a multi-shot science explainer, and a first-to-last-frame comparison with 1.1. 1
These are product claims and examples, not reported evaluation results. The post gives no measured figures for consistency, speech, lip-sync, camera control, reference adherence, or edit success rate, and no evaluation methodology. 1
Analysis: A polished demo shows an intended use; it does not establish how often outputs fail, how consistently a prompt reproduces across attempts, or how the model handles your assets. Treat the examples as use-case signals, not performance estimates.
How to evaluate a video AI model before you commit
Build a workflow that mirrors production: generate a draft, apply the edits you would actually make, and produce the intended delivery output. Test across the 3-to-15-second range the model supports, using reference images, motion clips, and audio cues as guidance inputs. 1
Log every attempt, not just the best one:
- Cost per accepted clip. Include drafts, failed attempts, edits, and any upscale step; the announcement gives no full-workflow price. 1
- Latency, measured separately for initial generation, follow-up edits, and final output. The post publishes no speed figures. 1
- Task completion and consistency. For product work, check whether the product stays consistent across frames; for character work, check appearance and motion. These are tests for capabilities Google describes, not results it reports. 1
- Edit success and locality. Check whether a requested change—such as swapping a background—preserves elements meant to stay fixed. Google describes this kind of edit but gives no success rate. 1
- Speech and lip-sync, if dialogue matters to your use case. The post identifies both as improvement areas but reports no measured results. 1
- Human acceptance, defined before testing and applied identically to every candidate.
Run the same prompts, references, duration and resolution targets, retry policy, and acceptance bar against alternatives. That comparison can help determine whether the lower price and longer clip improve your economics. If you are comparing several providers, a model catalog can help you shortlist candidates, and pricing that shows final per-response costs makes the cost-per-accepted-clip tally easier to keep.
The reasonable call from this announcement alone
The post supports a longer stated maximum clip—15 seconds, up from 10—a listed 720p rate of $0.07 per second, and a workflow involving iterative edits with optional upscaling. 1 It does not provide measured speed, full-workflow cost, comparative quality, or production reliability results. 1
Analysis: If your output fits in short clips and benefits from conversational refinement, the new duration ceiling and price make Gemini Omni 1.2 Flash a reasonable candidate for a controlled test. 1 For hard cost targets, high-volume generation, or quality-sensitive deliverables, evaluate workflow-level costs and task-specific performance before committing.
FAQ
What is the maximum clip length for Gemini Omni 1.2 Flash? Google says Gemini Omni 1.2 Flash can generate and edit clips from 3 to 15 seconds in one-second increments. The prior version supported clips up to 10 seconds. 1
How much does Gemini Omni 1.2 Flash cost? Google lists $0.07 per second for 720p video and says that is 30% lower than Gemini Omni 1.1. The announcement does not list prices for other resolutions or the full draft-edit-upscale workflow. 1
Does the announcement prove that Gemini Omni 1.2 Flash is faster? No. Google positions the model around speed, but the announcement gives no measured generation-speed figures, latency results, benchmark scores, or independent tests. 1
What video capabilities does Gemini Omni 1.2 Flash support? Google says users can guide generation with character, product, or style images, motion clips up to three seconds, and audio cues. It also describes conversational edits across full 15-second clips. 1
How should teams evaluate Gemini Omni 1.2 Flash? Test representative prompts and references across clip lengths, edits, resolutions, and retry scenarios. Measure cost per accepted clip, latency, consistency, edit success, and speech and lip-sync where relevant; the announcement does not provide those results. 1
Questions
- What is the maximum clip length for Gemini Omni 1.2 Flash?
- Google says Gemini Omni 1.2 Flash can generate and edit clips from 3 to 15 seconds in one-second increments. The prior version supported clips up to 10 seconds [1].
- How much does Gemini Omni 1.2 Flash cost?
- Google lists $0.07 per second for 720p video and says that is 30% lower than Gemini Omni 1.1. The announcement does not list prices for other resolutions or the full draft-edit-upscale workflow [1].
- Does the announcement prove that Gemini Omni 1.2 Flash is faster?
- No. Google positions the model around speed, but the announcement gives no measured generation-speed figures, latency results, benchmark scores, or independent tests [1].
- What video capabilities does Gemini Omni 1.2 Flash support?
- Google says users can guide generation with character, product, or style images, motion clips up to three seconds, and audio cues. It also describes conversational edits across full 15-second clips [1].
- How should teams evaluate Gemini Omni 1.2 Flash?
- Test representative prompts and references across clip lengths, edits, resolutions, and retry scenarios. Measure cost per accepted clip, latency, consistency, edit success, and speech and lip-sync where relevant; the announcement does not provide those results [1].
Sources
- Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips · Google Deepmind · Nov 5, 2026
How this article was made: an AI writer persona researched and wrote it from the sources listed above; separate AI passes edited it, checked its claims against those sources and audited it; it was published after passing automated editorial checks, without an individual editor's approval. How the Yelno blog works.