Veo 3 in 2026: What It Costs per Second and How to Run It

Key facts

• Model IDs: veo-3.1-generate-001, veo-3.1-fast-generate-001, veo-3.1-lite-generate-001
• Clip length: 4, 6 or 8 seconds (reference-image-to-video: 8 seconds only)
• Resolution: 720p or 1080p input; 720p, 1080p or 4K output on the Standard tier
• Frame rate and format: 24 FPS, mp4
• Prompts: English only
• Region: United States, us-central1
• Price per second: $0.05 to $0.60, depending on tier and resolution (as reported, November 2026)
• Audio: native, generated with the video

What Veo 3.1 can do

Google's Veo 3.1 model documentation lists text-to-video, image-to-video, first-and-last-frame generation, reference images, video extension, sound generation and C2PA content credentials. DeepMind's page adds style reference, camera controls, outpainting and adding an object to an existing shot.

Limits that matter for planning: up to 4 videos per prompt, an input image of no more than 20 MB, and 50 regional online prediction requests per minute. The standard model reached general availability on November 17, 2025, with a retirement date of May 4, 2027 or later.

Three model IDs, three price points, and the differences are real. Fast drops 4K output. Lite, per the same documentation, does not take image input at all, so it is text-to-video only.

Veo 3 pricing per second: every route in one table

There are four ways to pay for Veo 3, and they don't share a unit. Google's consumer plans sell monthly credits. The Gemini API sells seconds. Vertex AI (now Gemini Enterprise Agent Platform) doesn't sell seconds at all. Third-party hosts resell the same model at their own rate.

All figures in this table are as reported by US pricing trackers in November 2026. Google's Gemini API pricing page did not render when we checked, so the per-second rates come from round-ups published in September and October 2026, including Veo 3 Pricing in 2026: Every Way to Pay for Google Veo 3.1 (Sep 29, 2026) and Google Veo 3 Pricing 2026: API Costs and Access (Sep 28, 2026). Check Google's own page before you commit to a budget.

The third-party figures come from Cheapest Veo 3.1 API. $0.24 per 8s clip. (2026). That $0.241 rate is a reseller price, not Google's, and resale terms can change without notice. Treat it as a starting point to verify, not a promise.

Vertex is the route most round-ups skip. Google's own documentation for Veo 3.1 lists pay-as-you-go as not supported and Provisioned Throughput as supported. If you're on Google Cloud and want Veo per second, the Gemini API is the path, not Vertex.

What an 8-second Veo 3 clip actually costs

Per-second rates are the easy part. The number that lands in your invoice is per-second rate times seconds generated, and generated includes the takes you throw away.

One clip

An 8-second 1080p clip with audio costs $0.64 on Lite, $0.96 on Fast and $3.20 on Standard, at the reported November 2026 rates. Same prompt, same length, five times the price between the cheapest and most expensive tier.

Ten clips

Ten 8-second 1080p clips, all kept: $6.40 on Lite, $9.60 on Fast, $32.00 on Standard. That's 80 seconds of finished video.

A 60-second cut with retries

A 60-second video needs at least eight 8-second clips. In practice you generate more than you keep. Say you run two takes per shot and keep one: 16 generations, 128 seconds billed.

• Lite at 1080p: 128 x $0.08 = $10.24 for 60 seconds of usable footage
• Fast at 1080p: 128 x $0.12 = $15.36
• Standard at 1080p: 128 x $0.40 = $51.20
• Standard at 4K: 128 x $0.60 = $76.80

Google bills the seconds it generates. A take you delete because the hands warped is a take you paid for. That is the cost most Veo round-ups leave out, and it's the one that decides whether a project fits its budget.

Compare it to a platform that refunds failed runs and you're comparing two different pricing models, not two different prices. On LumozAI's pay-as-you-go pricing, failed or interrupted runs are refunded automatically. A bad take that completes is still billed everywhere, including here. The difference is the failed request.

How to call Veo 3 from the API

Veo 3.1 on the Gemini API is a long-running operation. You submit a job, get an operation name back, poll it, then download the file. The shape below follows Google's documented pattern for generateVideos: model ID, prompt, duration, aspect ratio, then poll the operation.

import os, time
from google import genai

client = genai.Client(api_key=os.environ["GEMINI_API_KEY"])

operation = client.models.generate_videos(
    model="veo-3.1-fast-generate-001",
    prompt="A slow push-in on a diner counter at dawn, coffee steam curling, a waitress laughs.",
    config={"aspect_ratio": "16:9", "duration_seconds": 8},
)

while not operation.done:
    time.sleep(10)
    operation = client.operations.get(operation)

video = operation.response.generated_videos[0]
video.video.save("clip.mp4")

Four things to know before you run it. The prompt must be English. Duration must be 4, 6 or 8. The model ID decides which tier you're billed at, so swapping veo-3.1-fast-generate-001 for veo-3.1-generate-001 multiplies your per-second cost by roughly four. And the 4K output only exists on the Standard model.

Vertex is a different setup. You need a Google Cloud project with billing enabled, and Veo 3.1 there is sold through Provisioned Throughput rather than pay-as-you-go, per Google's model documentation. Budget for reserved capacity, not a per-clip line item.

Veo is not in the LumozAI API reference or the model list. If you want to call it, you call Google.

The closest video models on LumozAI, priced against Veo

LumozAI doesn't host Veo. What it hosts is four video models on the same REST endpoint, priced per second or per run. Here's how they compare on the numbers.

Lumoz Video runs 4 to 15-second clips, takes a first frame and a last frame image, and supports 16:9, 9:16, 1:1, 4:3, 3:4 and 21:9. The Lumoz Video model page has the full parameter list. Draft mode bills at half rate for preview passes.

What Veo has that these don't: native dialogue and sound generated with the picture, and 4K output. Lumoz Video's documentation doesn't claim native dialogue audio, so plan to add sound separately if your cut needs it. What LumozAI has that Veo doesn't: a per-second rate under $0.07 at every resolution, and one key across video, image, audio and language models.

Pricing model matters as much as the rate. LumozAI is pay-as-you-go only. No subscription, top up any amount from $10, credits valid 365 days, failed runs refunded automatically, and the same balance works in the Studio and through the API. Google's consumer plans are the opposite: a monthly fee that buys Flow credits.

A working LumozAI video call

Video requests on LumozAI are asynchronous. You post to https://lumozai.com/api/v4/run with a bearer token and get a request_id back immediately. Then you either poll GET /api/v4/status/{id} or pass a webhook_url and let LumozAI POST to your server when rendering finishes. The Sora alternative API guide walks through the same pattern.

curl -X POST "https://lumozai.com/api/v4/run"   -H "Authorization: Bearer YOUR_LUMOZ_API_KEY"   -H "Content-Type: application/json"   -d '{"model_id": "lumoz-video", "prompt": "Slow dolly along a rain-slicked Tokyo alley at night, neon reflections", "duration": 8, "aspect_ratio": "16:9", "resolution": "768P", "webhook_url": "https://yourapp.com/webhooks/video"}'

That request costs $0.40 at $0.050/sec for 768p. The response carries a request_id:

{"success": true, "request_id": "req_8f2c1a9b"}

Python, with polling instead of a webhook:

import time, requests

API_KEY = "YOUR_LUMOZ_API_KEY"
headers = {"Authorization": f"Bearer {API_KEY}"}

job = requests.post("https://lumozai.com/api/v4/run", headers=headers, json={
    "model_id": "lumoz-video",
    "prompt": "Wide shot of a lighthouse in a storm, waves breaking, handheld camera",
    "duration": 8,
    "aspect_ratio": "16:9",
    "resolution": "768P",
}).json()

request_id = job["request_id"]

while True:
    status = requests.get(f"https://lumozai.com/api/v4/status/{request_id}", headers=headers).json()
    if status.get("status") in ("completed", "failed"):
        break
    time.sleep(5)

print(status)

Two costs to know: 1080p bills at $0.066/sec, and a 15-second 1080p clip is $0.99. A failed run refunds to your balance without a support ticket.

If you'd rather test before writing code, the Lumoz Video Studio playground runs the same model with the same parameters. And if you work inside an AI editor or Lumoz LLM, the LumozAI MCP connector exposes the video models as tools you can call from chat.

Which route to pick

Choose Veo 3.1 when the video needs spoken dialogue or synchronized sound generated with the picture, you want 4K output, and you can absorb $0.40/sec on Standard. At that rate a 60-second cut with one retake per shot runs about $51. That's a reasonable price for a hero ad and a bad price for 200 social clips.

Choose LumozAI when you need volume, you're iterating, or you want one key across video, image and audio. Lumoz Video at $0.033/sec is about 12 times cheaper per second than Veo 3.1 Standard at 720p or 1080p. An 8-second 768p clip is $0.40. Draft mode at $0.0165/sec makes a first-pass review cheap enough to run on every shot.

Most teams end up on both. Veo for the two shots that carry the campaign, Lumoz Video for the other forty.

Frequently asked questions

Create a LumozAI account, top up from $10, and run Lumoz Video in the Studio before you write a line of code. One key covers video, image, audio and language models, and failed runs come back to your balance.