Gemini Omni 1.2 Flash: 15-Second Video, 30% Lower Price, and General Availability

Google DeepMind released Gemini Omni 1.2 Flash on November 5, 2026. The headline changes: native video up to 15 seconds, multi-turn conversational editing, tighter reference following, and a price cut of 30% to $0.07 per second for 720p video. Google also says this is the first time Gemini Omni is generally available (Google's announcement).

This is a version update, not a new family. It follows Gemini Omni 1.1 Flash in August and the original Omni Flash launch in July. All numbers below come from Google's post unless we say otherwise. We could not open Google's pricing page content in our tooling, so check the Gemini API pricing page for your resolution before you budget.

Update (November 5, 2026): Vids rollout details

Google's Workspace Updates post for November 5, 2026 says Gemini Omni 1.2 Flash is now inside Google Vids. Users can slide a clip length from 3 to 15 seconds. Google lists better reference consistency, so characters and products look the same across generations, and a wider range of speaker voices. The page says the feature has no admin control. The full rollout starts on November 5 for Rapid Release and Scheduled Release domains, with feature visibility taking 1 to 3 days.

Availability covers Business Starter, Standard and Plus, Enterprise Starter, Standard and Plus, Education Plus, personal Google accounts, and Google AI Pro and Ultra.

The post does not state a free monthly allowance. It says users with the AI Expanded Access add-on get higher usage limits and links to Google's AI limits page, so check that page for the cap on your account before you plan work around it.

TL;DR: what changed

Longer clips: 10 to 15 seconds

Omni 1.1 topped out at 10 seconds per generation. Omni 1.2 Flash goes to 15 and lets you choose any length from 3 to 15 seconds. Google says that is enough for "complete stories," product explainers or a full social ad spot in a single generation.

For context, our Omni 1.1 coverage described 40-second extensions built by chaining. Native length matters because every extension step is a chance for drift. A single 15-second shot has fewer seams than three stitched 5-second ones.

Reference adherence and consistency

Google says you can combine character, product or style images, short motion clips of up to three seconds, and audio cues, and the model "holds onto those details from start to finish." Faces, garments, packaging typography and brand colors should stay steady across camera angles.

The post gives concrete demos rather than only claims:

• A virtual try-on sequence from one model photo plus front and back shots of three outfits, 15 seconds, with 360-degree turns and match cuts.
• A five-layer poster animation where Google says 1.1 matched the reference less closely than 1.2.
• A continuous FPV drone flight through a Pantheon-style dome from two reference photos, with no jump cuts.
• A first-frame to last-frame push from an exterior to an interior shot, shown side by side for 1.1 and 1.2.

These are Google's own picks. Treat them as best cases. Google itself notes that work continues on "fast sports action, rapid script changes and complex physics."

Multi-turn editing

The most practical change for production work is conversational editing. You refine one clip turn by turn: swap a product or background, change lighting or weather, or switch the aspect ratio, instead of regenerating. Google's example swaps the background of a clip five times, from a striped tent to a forest, a stable, a market fair and a castle, while keeping the characters' performances and camera framing.

Another example replaces a rippling octopus with a dancer whose gown mirrors the same fluid motion. Google calls this multi-turn motion transfer.

Why it matters for cost: a fix by follow-up edit should cost less than a fresh 15-second generation. Google makes that point itself: draft at low resolution, fix details with edits, then upscale the best result.

Speech, lip-sync and camera control

Google says dialogue is more natural, with clearer lip-sync and distinct multi-speaker conditioning. In its example, one scene image plus four character, mouth and one-second voice references route each voice to the right character. Camera work gets attention too: sweeping FPV flythroughs, zoom-outs and orbits with fewer unwanted jump cuts.

What it costs

Google's number is $0.07 per second at 720p, a 30% cut from Omni 1.1. Simple math from that figure (ours, not Google's):

Upscaling to 1080p or 4K is a separate step, and Google does not give those prices in the post. Our August write-up mentioned a draft-then-upscale model, but prices may have changed since, so use the live pricing page rather than our older numbers.

A sensible budget plan: generate drafts at 360p, pick the best, fix details with a follow-up edit, then upscale one final version. That keeps most spend on one clip rather than many full renders.

Where you can use it

The Vids line is notable. Free 15-second generation inside a Workspace tool puts pressure on standalone video apps for short business clips.

How it compares

Google does not publish benchmark scores in this post, and we will not invent any. For a cross-model view, our earlier Runway Aleph 2 vs Gemini Omni comparison shows how the family stacked up at I/O, and our I/O Omni piece covers the launch. For open alternatives, see Kandinsky 6 Video. A fair test is cost per usable second: run the same brief on each model and count how many generations you need before one is good enough.

How to try it

1. Open Google AI Studio, pick Gemini Omni 1.2 Flash and write a one-shot brief with a start frame and one reference image.
2. Generate at 360p. Check subject identity, text rendering and camera path.
3. Use one follow-up turn to fix a single detail. Avoid rewriting the whole prompt.
4. Upscale only the keeper.

A starter prompt:

15 seconds. One continuous shot, no cuts. Start on the product box from the reference image on a matte pink table.
Slow orbit left, then push in on the logo. Keep packaging text and brand colors exactly as in the reference.
Soft studio light. Quiet ambient audio.

The Omni timeline so far

Omni has moved fast. The original model arrived in May at Google I/O, and the Flash tier followed in July. Version 1.1 came in late August with longer extensions and first and last frame controls. Version 1.2 arrives about ten weeks later with longer native clips, lower cost and a general availability label.

That cadence tells you something about planning. A video pipeline built around one version's limits, such as a 10-second cap, can be outdated within a quarter. Keep clip length, resolution and model name in config, not hard-coded in prompts or scripts, so a version bump is a one-line change.

What to test before you commit

Google's examples show strengths, so build your own test set around the areas it admits are still weak and the areas that matter to your work.

• Text on packaging. Run five prompts with brand text and check spelling at 720p and after upscaling.
• Identity across turns. Make a clip, then apply four edits in a row. Compare the character's face on the first and last turn.
• Fast motion. Try a sports or action brief. Google lists this as unfinished work.
• Audio and lip-sync. If you use voice references, check that each voice lands on the right character in a two-person scene.
• Cost per keeper. Log how many generations and edits you need per accepted clip. That number, not the per-second price, is your real cost.

Record results in a simple table and re-run it when the next version ships. That gives you a repeatable way to judge updates instead of relying on launch demos.

Risks and open questions

• Deepfake and likeness risk. Better face consistency and lip-sync make convincing synthetic people easier. SynthID helps with detection, and our guide to Google's SynthID verification site covers how checking works. Watermarks can still be stripped by some edits, so do not treat a missing mark as proof of authenticity.
• Vendor-chosen demos. Every example in the announcement is Google's. Independent tests will show failure rates, especially on the weak areas Google names.
• Lock-in. Reference and edit workflows are tied to one model's behavior. A prompt that works on 1.2 may not transfer.
• Pricing details. Only the 720p rate is in the post.

What this means for what you build or pay

If you make short ads, product clips or explainers, 15 seconds in one generation plus cheaper edits lowers cost per finished clip. If you build apps on the Gemini API, general availability and Provisioned Throughput remove the main reason enterprises gave for waiting. If you already use Workspace, test Vids first because it carries no extra charge for 15-second clips.

Related reading

• Gemini Omni 1.1 Flash: the August update
• Gemini Omni Flash: the original launch
• Runway Aleph 2 vs Gemini Omni
• Google Flow and creative agents
• Google SynthID verification site

Details are accurate as of November 5, 2026. Prices, limits and availability can change, so confirm them on Google's pricing and docs pages.