Fifteen seconds changes the format, not the medium
Ten seconds is a clip. Fifteen seconds is closer to a complete unit of work: one product explainer, one social ad spot, one short scene with a beginning and an end. Google makes exactly that pitch, and on the arithmetic of duration it is fair. A creator who previously assembled a 15-second spot from multiple shorter generations can now attempt the entire sequence in one native generation. That removes the need to join separate clips in workflows where the model can produce the desired result in a single pass.
The detail worth keeping is that 15 is a ceiling, not a default. Native generation and editing now support durations from 3 to 15 seconds in one-second increments, with a 10-second default and 720p, 16:9 as the default output. Up to six reference images are supported per request.
Video extension is a separate operation: it appends 3 to 10 seconds per turn, up to 30 seconds in total. It cannot prepend or insert footage.
First-to-last-frame interpolation is also separate and remains limited to outputs of 3 to 10 seconds. Anyone quoting this release as a 30-second model has merged two different features.
The price, counted in tokens
The headline price survives scrutiny, with an important qualification: it is the price of video output, not the all-in cost of a finished clip. Google quotes $0.07 per second for 720p output, 30% below Omni 1.1's $0.10 rate. The underlying billing is token-based: $12 per million video-output tokens, at 5,792 tokens per second of 720p video. That works out to approximately $0.0695 per second, or $0.70 for 10 seconds and $1.04 for 15 seconds. These are our calculations from Google's published token rates, not guaranteed final-generation costs.
Input text, images and video are billed separately at $1.50 per million input tokens. The API has no free tier, and retries can incur additional generation charges. The actual cost of a finished clip therefore depends on its inputs, output duration, failed attempts and subsequent edits. Text output carries its own rate of $9 per million tokens.
The honest saving in this release may not be the 30% at all. Conversational editing, where a follow-up prompt swaps a background or a product while the rest of the clip stays put, replaces full regenerations with revisions. For a team producing ad variants, avoided regenerations are worth more than a cheaper second.
Access splits by surface, and the distinction matters for budgeting. The Gemini API is paid, with no free tier. Google Vids offers 15-second clips at no additional cost to Google account and Workspace subscribers, while the Gemini app rollout is limited to AI Ultra subscribers. Flow and Google Ads are additional distribution surfaces, but their availability and commercial terms should not be conflated with the API.
Existing Omni 1.1 API users also have a transition deadline. Google says version 1.1 will remain available during a four-week transition. On December 3, 2026, requests using the 1.1 model identifier will automatically route to 1.2 and receive the newer, lower rates, without requiring code changes. Teams that need to preserve the old model's behavior should test their workflows before that date rather than assume the transition is behaviorally invisible.
Consistency is the claim that needs a second source
Length and price are checkable. Consistency is not, yet. Google says 1.2 holds faces, garments, packaging typography and brand colors across camera angles, with less drift and morphing than 1.1, and supports that with its own side-by-side evaluations.
Google DeepMind's own side-by-side evaluations report a 62% visual-consistency win rate for Omni 1.2 against 1.1 in text-to-video, implying a 38% win rate for the older model. For multi-reference identity consistency, Google reports a 63% win rate against 37% for 1.1. Google has not published a sample size alongside these figures, and no independent evaluation was found in the sources checked for this article.
That does not make the claim false. It makes it a vendor measurement at launch, which is the normal state of a model release and the reason to test before migrating production work.
The wider Omni product supports a richer reference workflow, including character, product and style images, short motion clips and audio cues. Those capabilities should not be assumed to be available through every API route. The current Gemini API documentation does not support voice editing, uploaded audio inputs or audio references; Google Cloud separately says audio references require allowlisting. A team can run its own product shots through both versions in an afternoon and count the drift itself.
The limits Google published alongside the launch
The documentation is unusually specific about what the API does not do. It fully supports English and has not evaluated other languages. Voice editing, uploaded audio inputs and audio references are not supported in the current API, and Google Cloud's general-availability announcement places audio references behind allowlisting. Google's own post closes by saying work continues on fast sports action, rapid script changes and complex physics.
Google Cloud's GA announcement adds enterprise infrastructure, service-level agreements and Provisioned Throughput. Photoroom is already using the model for its Video Ad Maker and Product Animator, according to the same announcement. Google says generally available Omni 1.2 outputs carry an imperceptible SynthID watermark.
Sources
- [1] Google (primary), "Gemini Omni 1.2 Flash: 15-second video and lower pricing", blog.google, November 5, 2026 (read first-hand in full at Gate 1: general availability across the Gemini API and Gemini for business; native 15-second creation, up from 10 in 1.1, in one-second increments from 3 to 15; price dropped 30% to $0.07 per second for 720p; multi-turn conversational editing across full clips; reference adherence for faces, garments, packaging typography and brand colors; draft at 360p or 720p and upscale to 1080p or 4K; GA backed by Google Cloud with Provisioned Throughput and an imperceptible SynthID watermark; availability in AI Studio, Google Cloud console, Flow, Google Vids at no additional cost with a Google account or Workspace subscription, Gemini app rolling out to AI Ultra subscribers, and Google Ads; closing admission that fast sports action, rapid script changes and complex physics remain under development)Read source
- [2] Google Cloud (primary), "Gemini Omni 1.2 Flash now GA: cost-effective video", Google Cloud Blog, November 5, 2026 (read first-hand in full at Gate 1: GA on Google Cloud with enterprise-grade infrastructure and security and Provisioned Throughput; "Coming soon to Gemini for business" in its get-started list, where the main launch post describes business availability as part of the GA; audio references gated behind allowlisting; Photoroom powering its Video Ad Maker and Product Animator, with its CEO citing general availability and SLAs; $0.07 per second of 720p, 30% less than Omni 1.1 Flash)Read source
- [3] Gemini Developer API pricing page (primary; checked live at build on November 5, 2026, after an earlier crawl the same day still showed only the 1.1 section: the live page now lists Gemini Omni 1.2 Flash with input at $1.50 per million tokens (text / image / video), output at $9.00 per million text tokens and $12.00 per million video tokens, and no free tier; billing at 5,792 tokens per second of 720p video equates to approximately $0.07 per second, about $0.70 for a 10-second clip and $1.04 for 15 seconds; the 1.1 section, at $17.50 per million video tokens or about $0.10 per second, states that effective December 3, 2026 all API requests to gemini-omni-1.1-flash automatically route to 1.2 and are billed at 1.2 rates)Read source
- [4] Google DeepMind, Gemini Omni model page (read at Gate 1: content created or edited with Omni in the Gemini app, Google Flow or YouTube includes the imperceptible SynthID digital watermark and C2PA Content Credentials — that combined scope is stated for those surfaces, not for every API output)Read source
- [5] Unite.AI, "Google's Gemini Omni 1.2 Flash Reaches General Availability", November 5, 2026 (developer-documentation detail read at Gate 1: durations from 3 to 15 seconds with a 10-second default, 720p and 16:9 defaults with 9:16 supported, up to six reference images per request, extension appending 3 to 10 seconds per turn to a 30-second total with no prepend or insert, interpolation limited to outputs of 3 to 10 seconds; DeepMind side-by-side win rates of 62% visual consistency against 38% and 63% multi-reference identity consistency against 37%; API limits — English fully supported and other languages not evaluated, voice editing, uploaded audio inputs and audio references unsupported in the current API)Read source