Gemini Omni 1.2 Flash API and Pricing, Explained Through a Real Pipeline

Gemini Omni 1.2 Flash is a video model from Google and the successor to Gemini Omni 1.1 Flash. It takes a prompt, and optionally a first and last frame, and returns a clip with an audio track in 16:9 or 9:16. Two things changed from 1.1: clips now run from 3 to 15 seconds instead of 3 to 10, and the listed per-second rate dropped by about 30 percent at every resolution.

It reached OpenRouter on 5 November 2026. On 6 November every one of our five test requests failed on the provider side, so it stayed off the canvas for a day. On 7 November it rendered, and it went onto Treza's canvas and into chat the same morning. Every parameter and price below is read off Treza's live video model catalog as of 9 November 2026, and the behavior was checked with real renders.

What you get before you talk about price

The 15 second ceiling is the headline change. A 12 second product move or a 15 second establishing shot is one generation here, where 1.1 would have needed two clips and a stitch. We have rendered 3 second clips only so far, so the long end of that range is the listing, not a measurement. Capability detail lives on the Gemini Omni 1.2 Flash model page.

How Gemini Omni 1.2 Flash bills

Per second of output, at a rate set by the resolution. Audio is included at every tier:

One caveat, the same as with 1.1: the catalog lists the 1080p rate with no resolution label. It sits between 720p and 4K, and 4K is exactly twice it, so we read it as 1080p. We have rendered 360p and 720p only, so the 1080p and 4K rows are the listing and nothing more.

The steps between tiers are as steep as before: 720p costs three times the 360p rate, and 4K nine times. Draft low, then render the keeper once at the tier the placement needs.

Against Gemini Omni 1.1 Flash

The cut is the same at every tier:

A 10 second 720p clip that cost about $1.01 on 1.1 costs about 70 cents on 1.2, with the same inputs, aspect ratios and audio-always-on rule. Unless a render on 1.1 gave you something you need to match exactly, there is no price reason to stay on it. We wrote up 1.1 in Gemini Omni 1.1 Flash API and pricing.

Against Veo 3.1

At 720p with audio, Veo 3.1 Fast lists $0.10 per second and Veo 3.1 Lite lists $0.05. Gemini Omni 1.2 Flash at 720p sits between them at about 7 cents. Where 1.1 cost about the same as Veo 3.1 Fast, 1.2 undercuts it by about 30 percent, and its 360p tier at 2.3 cents per second is a draft price neither Veo model lists. Rates across the catalog are collected in the AI video cost breakdown.

What three renders actually measured

Before a model goes in the picker we run it through the node's own request shape and read the bill. Here that was three 3 second clips, with the same barista prompt we used on 1.1: steamed milk poured into a latte in a sunlit cafe, slow push-in, shallow depth of field, natural ambient sound.

Text to video at 360p, vertical. Back in 28 seconds at 360 by 640 pixels, 24 frames per second, with an audio track. A steel pitcher pouring into a dark cup, a rosetta forming, cafe shelves blurred behind. The charge was the listed 360p rate times three seconds, 7 cents.

Text to video with no resolution set, vertical. Back in 33 seconds at 720 by 1280, 24 frames per second. The same shot with more detail: the barista's apron, the espresso machine, steam off the pour. It billed the 720p rate, 21 cents, which is how we know 720p is the default.

A first frame with an unrelated prompt, at 360p. The still was a landscape product photo of a folded navy linen shirt on a light wooden table. The prompt was the barista. The clip came back in 35 seconds at 640 by 360, 7 cents. It opened on the shirt, and by the middle of the clip it was a hand pouring milk into a glass cup on a wooden counter.

Every charge matched the listing to the cent, and a first frame did not change the rate.

A first frame is a starting point, not an anchor

That third render is the one to remember, and it is where 1.2 behaves differently from 1.1.

On 1.1, a first frame with an unrelated prompt was ignored: the clip opened on a fresh cafe scene and the shirt never appeared. On 1.2 the clip did open on the photo, then left it for the scene the prompt described. Three seconds was enough to get from a folded shirt to a latte pour.

The working rule is the same for both models, and the Video Generation node carries a note for it: say what is in the image, then say how it moves. A prompt that describes the shirt and asks for sunlight drifting across it gives the model nothing to cut away to.

The first-frame render also came back 16:9: the node sends no aspect ratio when a frame is attached, and the model rendered landscape from a landscape still. For a vertical clip, start from a prompt, where 9:16 is a setting. The technique is covered in the first frame post, and the image to video generator covers it from the still's side.

About the audio

Every clip came back with an audio track, and there is no silent rate to choose instead: the catalog lists only with-audio prices for this model.

In all three clips the track was a quiet ambient bed, well below normal listening level, even though the prompt asked for natural ambient sound. The two 360p clips measured around minus 38 decibels mean volume, and we have not tested prompts that ask for dialogue or music. Treat it as room tone and add sound that carries in the graph: a Text to Speech node for narration, an Audio Generation node for music or effects, and the Sequence node to mix them over the picture and duck the clip's own audio under a voice.

What you actually pay on Treza

Gemini Omni 1.2 Flash through a Treza pipeline is metered from a prepaid credit balance. One credit equals $0.01, credit packs start at $5, credits never expire, and a failed generation charges nothing. That last rule did real work here: the five requests that failed upstream on 6 November would have cost a customer nothing.

Before a run starts, the platform estimates the media cost of the graph and refuses to start one your balance cannot cover. For this model the estimate follows the resolution the node is set to, and uses the 720p rate when none is set, because that is what an unset clip renders and bills. A cheap draft means setting 360p on purpose.

Calling Gemini Omni 1.2 Flash as an API without building the integration

On a Treza AI video pipeline, Gemini Omni 1.2 Flash is a node on a canvas. Set the duration, aspect ratio and resolution, wire it next to a script step, a voiceover step or a captioning step, and publish the whole chain as one versioned endpoint. Your application calls a typed /invoke route with JSON in and JSON out, or points an existing OpenAI SDK at the OpenAI-compatible route.

This release shows why that matters. Anyone who built on Gemini Omni 1.1 Flash on 6 November had a cheaper, longer-running successor the next morning, and on Treza that upgrade is a dropdown change and a republish. The endpoint your product calls does not change. Our hosted MCP server exposes the same pipelines as tools, so an agent in Claude or ChatGPT can render a brief and return the link.

Where Gemini Omni 1.2 Flash earns its place in a graph

Draft passes. Ten 3 second framings of one idea at 360p come to about 70 cents of provider cost. Render the one that works once at 720p or above.

Single long takes. A 12 or 15 second shot is one generation, with the sound of the scene already under it. We have not rendered at that length yet, so test one before you build a schedule on it.

Short vertical cuts. Text to video at 9:16 came back at 720 by 1280 in our test, with a quiet room tone a voiceover can sit on.

To see what it costs for your own work, open the AI video generator, pick Gemini Omni 1.2 Flash, set the resolution to 360p, and generate once.

Frequently Asked Questions

How much does Gemini Omni 1.2 Flash cost per second?

$0.0232 per second at 360p, $0.0695 at 720p, $0.1043 at 1080p and $0.2085 at 4K, audio included, as listed on the endpoint we run on 9 November 2026. A 3 second 360p clip is 7 cents and a 15 second 720p clip is about $1.04. Our 360p and 720p test renders billed exactly the listed rate.

How is Gemini Omni 1.2 Flash different from Gemini Omni 1.1 Flash?

It takes the same inputs and returns the same formats, but clips run up to 15 seconds instead of 10, and the listed rate is about 30 percent lower at every resolution. In our tests it also handled a first frame differently: 1.2 opened on the image before moving to the prompt's scene, where 1.1 skipped the image when the prompt did not describe it.

How long can a Gemini Omni 1.2 Flash clip be?

3 to 15 seconds in one generation, at any whole second. Our tests were 3 seconds. For pieces longer than 15 seconds, stitch generations with the Sequence node, which handles up to 12 shots, or assemble them in the timeline editor.

Can Gemini Omni 1.2 Flash start from an image?

Yes, it takes a first frame. Describe what is in the image in your prompt. In our test the clip opened on the photo and then cut to the scene the prompt described, because the prompt was about something else. The first-frame clip came back 16:9.

Can I call Gemini Omni 1.2 Flash through an API without a Google Cloud account?

Yes. On a Treza pipeline, Gemini Omni 1.2 Flash is a node you configure once and publish as a versioned HTTP endpoint, with a typed /invoke route and an OpenAI-compatible route. You can also ask for it by name in Treza chat. Generations are metered from prepaid credits that start at $5 and never expire.