Gemini Omni 1.1 Flash Costs $0.10 a Second for 720p Video

Gemini Omni 1.1 Flash is Google's video generation and editing model with native audio, and its price is now on the Gemini API pricing page: about $0.10 per second of 720p video. For a team that makes sprint demo bumpers or release videos, that puts a 10-second clip at roughly a dollar. Below: what the page says, what it leaves out, and where the model fits in a sprint.

What Gemini Omni 1.1 Flash costs per second of 720p video

The Gemini API pricing page lists gemini-omni-1.1-flash on the paid tier only:

• Video output: $17.50 per 1M tokens
• Input (text, image, video or audio): $1.50 per 1M tokens
• Text output: $9.00 per 1M tokens

A footnote says 720p output is billed at 5,792 tokens per second, which the page rounds to about $0.10 per second. Your prompt costs a fraction of a cent. The video output is the whole bill.

TOKENS_PER_SECOND_720P = 5_792        # pricing page footnote
USD_PER_MILLION_VIDEO_TOKENS = 17.50  # pricing page, video output

def clip_cost(seconds: float, takes: int = 1) -> float:
    return seconds * TOKENS_PER_SECOND_720P * USD_PER_MILLION_VIDEO_TOKENS / 1_000_000 * takes

print(f"${clip_cost(10):.2f}")           # one 10 s take
print(f"${clip_cost(10, takes=5):.2f}")  # best of five takes

This prints $1.01 and $5.07. The number to watch is takes, because prompt iteration multiplies the price (assuming every take is billed).

The page also lists no free tier and no Batch, Flex or Priority pricing for this model, so there is no bulk discount to plan around. Only the 720p rate is published. The docs also offer 1080p and 4k, which they describe as upscaled, so check your first higher-resolution invoice line before you budget for it.

Where Gemini Omni 1.1 Flash fits in a sprint

The models page describes it as fast video generation, editing, keyframe interpolation and extension with native audio. Clips run 3 to 10 seconds at 24 FPS. Per the Omni docs, extension adds 3 to 10 seconds per turn, up to 40 seconds in total.

Good fits are release-note bumpers, a title card for the sprint review recording, a 9:16 teaser for the announcement channel and a looping animation for the retro.

A bad fit is showing the product. A generated clip depicts a UI the model imagined, and a sprint review exists to show working software. Keep screen recordings for that.

import base64
from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-omni-1.1-flash",
    input="A paper rocket launches from a desk and flies over a city made of sticky notes, smooth continuous shot.",
    response_format={"type": "video", "resolution": "720p"},
)
with open("release-bumper.mp4", "wb") as f:
    f.write(base64.b64decode(interaction.output_video.data))

This goes through the Interactions API, not generate_content. The video comes back as base64 in interaction.output_video.data. The quickstart labels the Interactions API beta and its examples need a recent google-genai 2.x SDK, so upgrade before you test.

Gemini Omni 1.1 Flash as the Veo 3.1 replacement

The deprecations page lists veo-3.1-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-lite-generate-preview with a shutdown date of October 22, 2026. It names gemini-omni-1.1-flash as the recommended replacement for each.

Those dates are the earliest possible ones. Google says it will announce the exact date with advance notice. Nothing there says the Veo previews are off today, only that you should plan the move. If you ever used gemini-omni-flash-preview, it is on the same list, so move that too.

Omni is called through the Interactions API, so check how your current Veo call is wired. Expect a code change, not just a new model string.

Gotchas

• Paid tier only, with no Batch, Flex or Priority pricing listed.
• Editing or extending an uploaded video needs a clip of 10 seconds or less. Per the docs it is not available in the EEA, Switzerland or the UK, though editing a model-generated video is.
• store=false makes a clip uneditable in later turns via previous_interaction_id.
• Every video carries a SynthID watermark, invisible to viewers but detectable by software.
• System instructions, temperature and negative prompt fields are not supported. Write "Do not show X" in the prompt instead.

What to do on Monday morning

1. Search your repos and pipelines for the three Veo 3.1 model codes above.
2. Wrap generation calls in a cost helper like clip_cost and cap takes per clip.
3. Generate one test clip and compare the invoice line with your estimate.
4. Keep real screen recordings for the actual demo.

A per-second price makes video a line item you can forecast, but the Veo retirement window means the migration, not the clip cost, is what can bite you in production.