Gemini Omni 1.2 Flash is Google's video generation and editing model, made generally available on 5 November 2026. It extends clips from 10 to 15 seconds, lets you refine a clip over several conversational turns, and costs less than Omni 1.1. For product teams, general availability is the bigger change: video generation can now move from prototype into production.
What Google released
Gemini Omni 1.2 Flash was announced by Google on 5 November 2026 in Introducing Gemini Omni 1.2 Flash. The model ID is gemini-omni-1.2-flash. Google states that it is generally available on the Gemini API and Google Cloud, and that this is the first time Gemini Omni has been generally available.
Google Cloud followed with its own post, Gemini Omni 1.2 Flash is generally available, which adds that developers reach it through the Google Cloud console and the API, and that Provisioned Throughput is offered for reserved capacity. Business users get it in Google Vids now, with Gemini for business listed as coming soon.
The same model also reaches consumer and marketing tools: Flow, Google Vids, the Gemini app (rolling out to Google AI Ultra subscribers) and Asset Studio in Google Ads. Your marketing team may be using it before your product does.
What changed from Omni 1.1
Everything below comes from Google's announcement unless linked elsewhere.
- Longer clips. Generation and editing now run natively up to 15 seconds, up from 10 seconds with Omni 1.1. Duration is set in one-second steps from 3 to 15 seconds, so a full social ad or product explainer fits in one generation.
- Multi-turn editing. You can refine a clip turn by turn instead of regenerating it: swap a product or background, change lighting or weather, or switch aspect ratio. Edits work across the full 15-second clip.
- Reference following. A request can combine character, product or style images, short motion clips of up to three seconds, and audio cues. Google says faces, garments, packaging typography and brand colours stay consistent across camera angles. The Google Cloud post notes that audio references are gated behind allowlisting.
- Consistency. Less drift or morphing across motion and camera transitions, including when interpolating between a first and last keyframe or extending a clip.
- Speech and camera. More natural dialogue, clearer lip-sync and multi-speaker conditioning, plus camera moves such as flythroughs, zoom-outs and orbits with fewer unwanted jump cuts.
- Draft, then upscale. Drafts render at 360p or 720p, and finished clips can be upscaled to 1080p or 4K.
Google is open about the limits: it is still working on fast sports action, rapid script changes and complex physics. If your use case depends on any of those, test it first.
What it costs
Google prices the model per second of generated 720p video, at a lower rate than Omni 1.1. The Google Cloud post says savings are similar at other resolutions, but neither announcement gives the rates for 360p, 1080p or 4K. Check the Model Garden pricing page or the AI Studio pricing page for your route before you quote a cost per clip.
Because billing is per second of video, a full 15-second 720p clip is charged for all 15 seconds before any upscaling. Every edit turn and every retry is billed as more generated video, so the cost per finished clip depends on how many turns your users take, not on the rate card alone.
What it means for teams building AI products
CodeDTX's view: budget per finished clip, not per request, and design the edit loop before you design the prompt.
Make drafts cheap by default. Google's own workflow is to draft at a low resolution, fix details with follow-up edits, and upscale only the clip you keep. Build that into your product: default previews to 360p or 720p, and make upscaling an explicit step a user or reviewer chooses.
Cap turns and seconds per user. Multi-turn editing is useful, but an open-ended loop is an open-ended bill. Set a maximum duration, a maximum number of edit turns and a daily allowance per user or workspace, and show the remaining allowance in the interface. Our guide to controlling AI agent running costs covers the same budgeting pattern for agents.
Store the edit history. Keep each turn's prompt, references and output so a user can step back to an earlier version, and so you can explain later how a published clip was made.
Treat reference images as rights-managed input. Product shots, faces and brand assets you upload shape the output. Record who supplied each reference and whether you have the right to use it, and keep a person approving anything that goes public, as in our guide to human-in-the-loop review without friction.
Keep the SynthID watermark in your disclosure plan. Outputs carry an imperceptible SynthID watermark. That helps provenance, but it is not a visible label. If your market or your clients expect AI-generated video to be marked, add your own disclosure in the player or the metadata.
Plan for audio references separately. Audio cues need allowlisting on Google Cloud, so do not promise voice-matched clips in a product until your project has that access.
How to test it before you switch
- Build a fixed brief set. Take real briefs from each use case: product explainers, social ads, talking-head clips. Run them on your current model and on Omni 1.2 Flash at the same resolution, then compare. Our post on what AI agent evals catch covers keeping a test set useful.
- Count turns, not just outputs. For each brief, record how many edit turns it took to reach an acceptable clip and the total seconds generated. That number drives cost.
- Check consistency where it matters. Look closely at product packaging, logos, faces and on-screen text across camera moves, since those are the details Google says improved.
- Test your weak spots. Fast action, fast dialogue changes and complex physics are the known gaps.
- Keep your current model behind a flag while you compare, and plan for the day a model is deprecated.
When not to switch yet
Wait if your clips depend on fast sports action or complex physics, if you need audio-referenced output and have no allowlist access, or if your budget needs confirmed 1080p or 4K rates that you have not yet seen on the pricing page. If your current clips are 10 seconds or shorter and already pass review, the lower price is a reason to test, not a reason to rush.
Frequently asked questions
What is Gemini Omni 1.2 Flash?
Gemini Omni 1.2 Flash is Google's video generation and editing model, made generally available on 5 November 2026 with the model ID gemini-omni-1.2-flash. It creates and edits clips of 3 to 15 seconds, follows image, motion and audio references, and supports multi-turn editing. It is available on the Gemini API, Google Cloud, Flow, Google Vids, the Gemini app for Ultra subscribers and Google Ads.
How is Gemini Omni 1.2 Flash different from Omni 1.1?
Clips can now run up to 15 seconds instead of 10, with duration set in one-second steps. Users can refine a clip over several turns instead of regenerating it, references hold more consistently across camera moves, and speech and lip-sync are clearer. Google has also lowered the price compared with Omni 1.1, and this is the first generally available release of Gemini Omni.
How much does Gemini Omni 1.2 Flash cost?
Google publishes a per-second rate for 720p video and says it is lower than Omni 1.1. The announcements do not give rates for 360p, 1080p or 4K, so check the Model Garden or AI Studio pricing page for your route. Each edit turn and retry generates more billable video, so measure the seconds generated per finished clip on your own briefs before setting user limits.
Can I use Gemini Omni 1.2 Flash in production?
Yes. Google states that it is generally available on the Gemini API and Google Cloud, backed by Google Cloud's security, with Provisioned Throughput available for reserved capacity. Before launch, set duration and turn limits per user, keep an edit history for each clip, confirm you have rights to every reference image, and decide how you will disclose AI-generated video beyond the built-in SynthID watermark.
