Gemini Omni 1.2 vs 1.1: What Changed, Tested on Real Clips

Gemini Omni 1.2 Flash is Google's third-generation Omni video model: it renders clips with synchronized sound from text, images or video, and its headline change over Omni 1.1 is simple to state. Longer takes for less money. Google released it on 5 November 2026; we put it into production on BananaBanana four days later and have been running it next to 1.1 ever since.

Quick answer: Omni 1.2 renders 3 to 15 seconds in one take (1.1 stops at 10) and costs roughly 30% less per second at every resolution: $0.07 a second at 720p instead of $0.10. Sound is still always on, extension still tops out at 40 seconds, and a clip that animates between a first and a last frame is still capped at 10 seconds. For new work there's little reason to stay on 1.1.

Below are the numbers, the same prompt rendered on both models, and the cases where 1.1 still makes sense.

What changed between Gemini Omni 1.1 and 1.2?

Two numbers moved, and a handful of quality claims came with them. According to Google's announcement, the clip length goes up to 15 seconds "in one-second increments from 3 to 15 seconds", 720p drops to $0.07 per second, and the model gets clearer lip-sync, distinct multi-speaker conditioning and "significantly fewer unwanted jump cuts". The model card adds that outputs still carry an invisible SynthID watermark.

Here's how the two generations compare on our platform today:

One detail from our own billing logs: 1.2 spends exactly as many output tokens per second as 1.1 (5,792 per second at 720p, 17,376 at 4K). The model isn't doing less work. Google simply cut the token price from $17.50 to $12 per million, and the per-second saving falls straight out of that.

Gemini Omni 1.2 Flash, 10 seconds at 720p, sound on. Prompt in short: a bike mechanic tightens a bolt, looks up and says "Most squeaks come from one loose bolt. Check this one first, then ride." Cost: $0.70.

Same prompt on both models: how does 1.2 compare?

I ran one prompt through both generations with identical settings: 16:9, 720p, exactly 10 seconds. A woman with short grey hair in a denim apron, a bicycle workshop, one spoken line, "single continuous shot, no scene cuts", and an "Audio:" line asking for her voice, wrench clicks and room tone with no music.

The same prompt on Gemini Omni 1.1 Flash, 10 seconds at 720p. Cost: $1.00.

What I saw, frame by frame:

• Both kept it to one shot. A scene-detection pass with ffmpeg found zero cuts in either file. On 1.1 that instruction mattered a lot; without it Omni tends to edit a clip into several angles on its own.
• Both spoke the line. Each woman delivers the scripted sentence to camera with the voice generated in the same pass. The 1.2 track came out noticeably louder (about 4 dB higher average level), which is handy if you post without a mix.
• 1.1 followed the wardrobe more literally. It gave her the cropped grey hair from the prompt; 1.2 went for a chin-length bob. Small, but it's a reminder that "better on average" doesn't mean better on every take.
• 1.2 used the wrench as a prop. She holds it up toward the lens while she talks, which reads more like a real creator video than the 1.1 version.

Is that a benchmark? No. It's one pair of clips, and Omni gives you a different take every time (there's no seed to lock a composition). What it does show is that 1.2 didn't lose anything 1.1 was good at, and it cost $0.30 less for the same 10 seconds.

Does a 15-second take actually matter?

More than the five extra seconds suggest. Ten seconds is long enough for a hook, but it's short for a complete action: a pour from start to finish, a product being opened, a short review to camera. On 1.1 you'd render 10 seconds and then extend.

Gemini Omni 1.2 Flash, 15 seconds at 720p in a single take. Prompt in short: a ceramicist shapes a bowl on the wheel, smooths it with a wooden rib, cuts it free with a wire and lifts it onto a board; wheel hum and wet clay as audio, no music. Cost: $1.05.

I honestly expected this one to fail halfway. Omni has a habit of dropping later events in a prompt ("then she cuts it free…") and handing back a pretty but static scene. Not here: the hands shape the walls, the rib comes in, around the 11-second mark the wire goes under the base, and by second 14 the bowl is lifted off the wheel. Four steps, one take, no cuts. One clip proves little, but this is exactly the kind of multi-step prompt that earlier Omni versions kept flattening into a static scene for us.

Now the money. Getting 15 seconds out of 1.1 means a 10-second clip ($1.00) plus a 5-second extension, and an extension is billed on the full resulting length, so that second step costs 15 × $0.10 = $1.50. Total: $2.50. The same 15 seconds on 1.2 in one take: $1.05. That's less than half, and you skip a step where the extension could drift away from the original look.

How much does Gemini Omni 1.2 cost per clip?

Every second is billed by resolution, sound included:

Google lists 1.2 at about $0.0232, $0.0695, $0.1043 and $0.2085 per second for the four resolutions. We round to the cent, and at 4K we round down to $0.20.

Top-up bonuses lower the effective rate further. A crypto deposit of $50 or more adds 5% to the balance, $100 or more adds 10%, and an active promo code adds another 10% of each deposit on top. So $100 in crypto with a code becomes $120 of balance, which puts a 720p second at roughly $0.058 and a full 15-second 720p clip at about $0.88. Card, PayPal and SEPA payments are credited without the deposit bonus.

My workflow: draft at 360p ($0.30 for a full 15 seconds), fix the timing and camera there, then re-render the keeper at 720p. Remember the draft and the final are separate generations, so the final won't be a frame-for-frame upscale of the draft. Full tables for every model are on the pricing page.

What didn't change: the limits of Omni 1.2

The 15-second ceiling doesn't apply everywhere, and that's the part people trip over first.

• First and last frame: still 10 seconds. When we tested a 12-second clip between two keyframes, Google's API rejected it with "Interpolation duration 12 seconds exceeds maximum duration 10 seconds". The studio and the MCP server now cap that case at 10 seconds before charging, so you won't hit the error yourself.
• Extension appends 3–10 seconds per step, 40 seconds in total. Starting from 15 just gets you there in fewer steps. Nothing can be inserted at the start or in the middle.
• Sound can't be switched off. Every Omni clip has an audio track. If you need silence, Veo 3.1 renders without sound, and Wan 3.0 can switch it off for free.
• No seeds, no negative-prompt field, one sample per request. Write exclusions into the prompt ("no music", "no dialogue"); they work most of the time, not always.
• Speech editing is restricted. Per the model card, editing a video can change what people say, and Google is limiting that capability for now.

Google itself says work continues on fast sports action, rapid script changes and complex physics. My take: if a shot depends on a ball bouncing correctly, render a 360p draft first.

Should you switch from Omni 1.1 to 1.2?

For new projects, yes. Same features, longer takes, about a third cheaper at every resolution. I can't think of a fresh brief where 1.1 wins on paper.

Keeping 1.1 makes sense in two cases. You're continuing a series shot on it and want the look to stay consistent between episodes. Or you've tuned a set of prompts on 1.1 and its take on them is exactly what you want (the hair test above is a small example of how the two read a prompt differently). Both generations stay available in the generator and over MCP, and you pick per clip.

There's also a date to keep in mind. Google's release notes say that from 3 December 2026 the Gemini API will route requests for gemini-omni-1.1-flash to 1.2. Here both stay selectable for now; if that changes on our side, this post will be updated.

For a deeper look at what the Omni API accepts and refuses, see our Gemini Omni Flash API review. The model page with more sample clips is Gemini Omni 1.2.

FAQ

Is Gemini Omni 1.2 better than 1.1?

On the specs, yes: 15-second takes instead of 10 and about 30% lower prices, with the same features. Google also claims clearer lip-sync and fewer unwanted jump cuts. In our side-by-side test both handled a spoken line in one shot; 1.2 cost $0.70 against $1.00.

How long can a Gemini Omni 1.2 video be?

3 to 15 seconds in one generation, set to the second. With both a first and a last frame the limit is 10 seconds. Longer scenes come from extension, 3–10 seconds per step, up to 40 seconds in total.

How much does a 10-second Omni 1.2 clip cost?

$0.70 at 720p, $0.20 at 360p, $1.00 at 1080p and $2.00 at 4K, with sound included. Failed generations are refunded automatically.

Can I still use Omni 1.1?

Yes. Omni 1.1 Flash stays available next to 1.2 in the generator and the MCP server, and you choose the generation for each clip. Google plans to route Gemini API calls for 1.1 to 1.2 from 3 December 2026.

Can Omni 1.2 make a video without sound?

No. Every Omni clip has an audio track. You can describe it on an "Audio:" line or ask for "no music, no dialogue", but for a truly silent clip use Veo 3.1 or Wan 3.0.