Google's Veo 3 is one of the most discussed video models around, mostly because it does something few others do well: it generates audio alongside the video. Dialogue, sound effects, and background noise, all matched to what is happening on screen, in the same request. If you are deciding whether to build on it, this guide covers what it does, what it costs, how to get access, and how to call it from code.
What Veo 3 Is
Veo 3 is Google's flagship video generation model, available through the Gemini API and Vertex AI. It takes a text prompt, or a prompt plus a reference image, and produces a short clip of up to 8 seconds at up to 1080p.
Worth knowing before you start: Google has since shipped Veo 3.1, which is also available on Apiframe. Everything below applies to the Veo 3 line generally, but check which version you are calling, since the newer one is usually the better default for a fresh build.
What sets it apart is that it makes the sound itself. Several models still hand you silent video that you then have to score, voice, and sound-design separately, so a prompt about a street musician playing violin in the rain comes back with the violin actually playing, rain hitting pavement, and city noise behind it, all lined up with the picture. Veo 3 was first to make this genuinely good, though it is no longer alone: Kling 3.0 and Seedance 2.5 now generate sound too. For which models do and do not, see our guide on APIs that support synchronized audio with generated video.
Google positions Veo 3 at the premium end of the market against Kling 3.0 and Seedance, and audio quality is the main reason it holds that spot. It is neither the cheapest nor the fastest, but for anything where sound matters, including dialogue-driven shorts, product demos with voiceover, and ads, it is usually the first one developers try. If you are comparing video models more broadly, our AI video API guide is a good place to start.
Key Features
It generates the audio
The headline feature. Veo 3 produces matched dialogue, sound effects, and background audio in the same pass as the picture, with no separate speech or sound design step. You can also switch audio off if you only need silent footage, which is useful when you plan to add your own voiceover or music later. Turning it off cuts the cost by about a third.
Resolution, length, and shape
Veo 3 outputs 720p or 1080p, in clips of 4, 6, or 8 seconds. The shape is configurable, so you can target landscape, square, or vertical depending on where the video is going. Eight seconds is a real constraint worth planning around, since it is shorter than most rivals, and the way to go longer is to generate several clips and join them.
Prompt following and camera direction
Veo 3 follows detailed prompts closely, including camera instructions like slow pan, tracking shot, or close-up, and the language of filmmaking generally. It also accepts negative prompts, so you can state what you do not want in frame.
Starting from an image
Rather than generating purely from text, you can pass a reference image as the opening frame and have Veo 3 animate from there. Useful for product shots, brand assets, or anywhere the first frame has to match something exactly. Our image-to-video tutorial walks through that pattern in code.
Pricing
Going direct through Google
Google bills per second of output, with the rate depending on the tier and whether audio is included. Recent public pricing puts Lite near $0.05 per second at 720p, Fast near $0.10, and Standard near $0.40 for silent output at either resolution. Adding audio raises that by roughly half, so an 8-second Standard clip with sound lands around $4.80. Google adjusts these tiers as new versions ship, so check the current Gemini API and Vertex AI pricing pages before planning a production budget. For how this compares to other models, see our AI video API pricing guide.
Fast against Standard
Fast trades some quality and prompt accuracy for a lower rate and quicker turnaround, which suits prototyping and high-volume, lower-stakes work. Standard is what most teams move to for production, particularly when the audio has to hold up. At four times the price, it is worth drafting on Fast and only finalizing on Standard.
Costs that catch people out
Three things surprise teams. Discarded attempts still cost money, so budget for the takes you throw away rather than only the ones you keep. Storage and data transfer add up if you are pulling large volumes of video out of cloud storage. And the gap between silent and audio output is easy to forget when estimating. Our guide to estimating monthly video costs works through the real figure at volume.
Getting Access
Straight from Google
Veo 3 is available through the Gemini API and through Vertex AI, Google Cloud's machine learning platform. Both need a Google Cloud account with billing switched on, and Vertex AI is built around usage caps and regional availability, so confirm it works in your region before committing to it as your only route.
Through a platform that bundles models
If you would rather not manage a Google Cloud integration, or you want to compare Veo 3 against other models without building twice, a single API across models is the other route. Apiframe exposes Veo 3 through one video endpoint alongside Kling, Seedance, Runway, Hailuo, and others, so switching is a change of one field rather than a second set of credentials and request formats. Our Kling 3.0 guide, Runway guide, and Seedance 2.5 guide cover setup for the alternatives.
Which route fits
If you already run on Google Cloud and only ever plan to use Veo 3, going direct through Vertex AI keeps things simple and avoids a middle layer. If you want to test Veo 3 against other models, prefer one bill to several, or would rather skip waiting lists and usage cap requests, a platform that bundles models gets you started faster.
Code Example
Here is a Veo 3 request through Apiframe's video API, which puts Veo 3 behind the same POST /v2/videos/generate endpoint as every other video model. Our getting started guide covers getting a key if you need one. Note that an 8-second Standard clip with audio costs real money, so start short while you are testing.
import requests
response = requests.post(
"https://api.apiframe.ai/v2/videos/generate",
headers={
"X-API-Key": "afk_your_api_key_here",
"Content-Type": "application/json",
},
json={
"prompt": "a street musician playing violin in a rainy alley, cinematic",
"model": "veo-3",
"veoParams": {
"duration": 8,
"generate_audio": True,
"resolution": "1080p",
},
},
)
data = response.json()
print(data) # {"jobId": "...", "status": "QUEUED"}The request comes back immediately with a jobId and a QUEUED status, since generation runs in the background. You either check the job yourself or supply a webhookUrl and let the API tell you when it is done. Checking looks like this.
import time
job_id = data["jobId"]
while True:
job = requests.get(
f"https://api.apiframe.ai/v2/jobs/{job_id}",
headers={"X-API-Key": "afk_your_api_key_here"},
).json()
if job["status"] == "COMPLETED":
print(job["result"]["videoUrl"])
break
elif job["status"] == "FAILED":
print(job["error"])
break
time.sleep(3)When the job finishes, result.videoUrl points to an MP4, hosted for 90 days, so download anything you intend to keep. A few settings on the veoParams object are worth knowing: duration takes 4, 6, or 8 seconds, resolution is "720p" or "1080p", generate_audio defaults to true and can be set false for silent output, and image takes a reference image URL to animate from. To skip checking manually, add a webhookUrl and webhookEvents: ["completed", "failed"] to the request body and the API will call your server instead, as our walkthrough on callbacks explains.
One caution on retries. If a request times out, do not simply send it again. A timeout means the response never reached you, not that the job never started, so a blind retry can run a second generation and bill you for both. At Standard rates with audio that is nearly $5 wasted. Send an idempotency key so the provider recognizes the repeat.
How It Compares
| Model | Approx. price per second | Max clip | Generates audio | Speed |
|---|---|---|---|---|
| Veo 3 (Standard) | $0.40 silent, ~$0.60 with audio | 8s | Yes | Moderate |
| Veo 3 (Fast) | ~$0.10 | 8s | Yes | Fast |
| Kling 3.0 | $0.24 standard to $0.49 pro with audio | 15s | Yes | Moderate |
| Seedance 2.5 | Lower, varies by resolution | 30s | Yes | Fast |
| Hailuo 2.3 | Flat per clip, from $0.41 | 10s | No | Moderate |
Rates shift often as providers adjust tiers, so treat these as rough positioning rather than budget figures. The practical read: Veo 3 sits at the premium end, and what justifies it is audio quality rather than price or speed. If cost per second decides it, Kling and Seedance come in lower. If clip length decides it, Veo 3's 8-second ceiling is the shortest here and Seedance 2.5 is nearly four times longer. If well-matched audio decides it, Veo 3 is still the one to test first. Our comparison of the best AI video generation APIs covers the wider field.
One name you will not find in that table: OpenAI discontinued the Sora API on September 24, 2026, with no replacement named, so it is no longer an option to compare against. Veo 3 is one of the usual destinations for teams that were building on it, as our post on Sora API providers explains.
What People Build With It
Marketing teams use Veo 3 for short ads and social content where a voiceover or matched sound effect adds real polish without a separate audio pass, which our guide to video APIs for mobile ad creatives covers in depth. Product teams use it for demo videos where narration has to line up with what is on screen. It turns up in storyboarding too, where directors want a quick sense of how a scene will look and sound before committing to a shoot. And because it handles speech natively, it suits short narrative pieces built around a character actually talking rather than silent footage with captions over it.
FAQ
Is there a free tier or trial credit?
Google periodically offers free credits through the Gemini API and Google Flow for testing, though the allowance changes. Apiframe gives new accounts 100 credits, which is worth setting expectations on: Veo 3 is a premium model, so that covers testing rather than production. Check current offers before picking a plan.
What are the rate limits?
That depends on your route. Gemini API and Vertex AI limits are tied to your Google Cloud allowance and project tier. Apiframe applies a flat 500 requests per minute per account whichever model you call. For video, the cap on how many jobs run at once usually matters more than the request limit.
Can I use the output commercially?
Terms are set by Google and vary by tier, and if you go through another platform, that platform's terms apply on top. Check the current terms for your specific plan before shipping generated video into a commercial product.
How does the price compare to Kling and Seedance?
Veo 3 Standard is the most expensive of the three per second, with Kling roughly half and Seedance lower still. The gap narrows if you use Veo 3 Fast, which is close to Kling on price. Whether the premium is worth it comes down to whether you need the audio quality, since all three now generate sound.
How do I make something longer than 8 seconds?
Generate several clips and join them, which is the usual approach but brings a catch: every join is a point where a face shifts or the lighting jumps, and matching audio across the seam is harder still. If length matters more than audio quality, a model that produces longer clips in one pass, such as Seedance 2.5 at 30 seconds, will give you a cleaner result than stitching Veo 3 output together.
Should I use Veo 3 or Veo 3.1?
For a new build, start with 3.1, since it is the newer release and available through the same routes. Veo 3 remains a reasonable choice if you have already built on it and are happy with the results, as the request shape is largely the same either way.
Apiframe Team
The team behind Apiframe - making AI generation accessible to everyone.