Veo 3.1 API Key: Free Access, Limits and Rate Limit

If you searched for a free Veo 3.1 credential, here is the short answer: Google does not offer one. On the Gemini API pricing page, every Veo 3.1 variant shows "Not available" in the free tier column, and the paid tier bills for each second of video you generate. That single line explains most of the confusion online, including the forum threads where students on a free Google AI Pro plan can chat with Gemini all day but hit a wall the moment they call Veo from code.

This article lays out what Veo 3.1 costs per clip, how Google's rate limits really work, how to stop 429 quota errors, and the cheapest ways to test the model before you spend money. Prices and limits come from Google's own documentation, checked on October 6, 2026. They move often, so confirm them on the official pages before you commit a budget.

What Google Actually Offers

Three Model Names to Know

Through the Gemini API, Veo 3.1 arrives as three preview model codes:

• veo-3.1-generate-preview is the standard model.
• veo-3.1-fast-generate-preview is the faster, cheaper tier.
• veo-3.1-lite-generate-preview is the lowest price option, limited to 720p and 1080p.

Every call is asynchronous. You submit a request, receive an operation object, and poll it until the clip is ready. Google lists latency from 11 seconds at best to 6 minutes during peak hours, so build your code around waiting instead of blocking.

You can try the same three models without touching Google Cloud: Veo 3.1, Veo 3.1 Fast and Veo 3.1 Lite are all available on Picasso IA.

Output Rules That Change Your Bill

The spec sheet matters because several options force a specific duration or resolution, and those choices set your price.

A few rules hide in the fine print:

• 8 seconds is mandatory for 1080p, 4K, reference images and extensions. A 4 second 1080p clip does not exist.
• Extensions add 7 seconds per step, up to 20 times, and run at 720p only.
• Each request returns one video, and the text prompt tops out at 1,024 tokens.
• Every clip carries a SynthID watermark.
• Google keeps generated files on its servers for 2 days.

💡 Tip: Download each clip the moment the operation finishes. After two days the file is gone, and you pay again to regenerate it.

Is Free Access Real?

What the Free Tier Does

Google's free tier is a no cost sandbox for text models. Veo is not in it. The pricing page lists Veo 3.1, Veo 3.1 Fast and Veo 3.1 Lite as "Not available" on the free tier, so a project without a billing account cannot run them.

That means any site promising a free Veo 3.1 API is usually reselling access through its own paid account, capping you hard, or collecting your email. Be suspicious of free access tokens from unknown sellers. No legitimate free tier exists to hand out.

Consumer Plans Are Not the API

Google AI Pro and Google AI Ultra subscribers can generate Veo 3.1 clips inside Google's own apps. That is a separate product with its own caps, and it does not give you API access. Forum posts show students on a free Google AI Pro plan hitting 429 RESOURCE_EXHAUSTED errors when they call Veo from code, even though text generation works fine for them.

If your goal is to see what Veo 3.1 produces and test prompts, an app or a browser tool is enough. If your goal is automation, you need a project with billing attached. The Veo 3.1 page on Picasso IA describes the model as free and online with no coding needed, which makes it a decent sandbox before you pay per second.

Pricing Per Second, Clip by Clip

The Price Table

Google bills the paid tier per second of generated video. These are the listed prices in US dollars, next to what one 8 second clip costs:

Two gaps stand out. Fast at 1080p costs $0.12 per second against $0.40 on the standard model, a 70 percent cut. Lite at 1080p costs one fifth of the standard price for the same resolution. A 4 second clip on the standard model lands at $1.60, so shorter runs add up slowly but they still add up.

Budget Math for 100 Clips

Take 100 clips of 8 seconds each, which is 800 seconds of video:

• Veo 3.1 at 1080p: 800 × $0.40 = $320
• Veo 3.1 Fast at 1080p: 800 × $0.12 = $96
• Veo 3.1 Lite at 720p: 800 × $0.05 = $40

Now a smarter plan. Draft all 100 ideas on Lite at 720p for $40, pick the 10 best, and re-render only those on the standard model at 1080p: 80 seconds × $0.40 = $32. The total is $72 instead of $320, about 78 percent less for the same ten finished clips.

💡 Tip: Prices apply to seconds of generated video. Before a large batch, check your billing dashboard to see how filtered or failed requests show up.

Rate Limits and Quota Errors

How Google Counts Limits

Gemini API limits are measured three ways: requests per minute (RPM), input tokens per minute (TPM) and requests per day (RPD). Exceed any one of them and the call fails. Two details catch people out:

• Limits apply per project, not per API credential. A second credential inside the same project gives you no extra room.
• Daily quotas reset at midnight Pacific time.

Your usage tier decides how much room you get:

Google also applies spend caps inside a rolling 10 minute window on the paid tiers, which matters when a script fires dozens of 4K renders at once.

Here is the honest part: Google's rate limit page does not publish a Veo specific RPM, concurrency or daily cap. Third party blogs quote figures such as 10 requests per minute for video, but I could not match them to any Google page, so treat them as rumors. Read your live numbers in the usage view of Google AI Studio instead.

Fixing 429 RESOURCE_EXHAUSTED

The message reads: You exceeded your current quota, please check your plan and billing details. Google staff describe it as a rate limit hit, meaning too many requests per minute. Their advice is to stay inside the model's limits and request a quota increase when you need more.

Work through this list in order:

1. Attach billing. A free tier project cannot call Veo at all, and the error looks the same.
2. Submit one job at a time. Poll the operation you already have instead of resubmitting.
3. Back off exponentially. Wait 1, 2, 4, 8 seconds before each retry, plus a little random jitter.
4. Request more quota once you are on a paid tier and still hitting the ceiling.

Here is a small retry wrapper for the official Python SDK:

import random
import time
from google.genai import errors

def submit_with_backoff(make_call, tries=5):
    for attempt in range(tries):
        try:
            return make_call()
        except errors.APIError as exc:
            if exc.code != 429 or attempt == tries - 1:
                raise
            time.sleep(2 ** attempt + random.random())

Copy Prompt

Three Ways to Get Access

Gemini API Through AI Studio

This is the direct route. Create a Google Cloud project, attach billing, generate a credential in Google AI Studio, then install the SDK with pip install google-genai. This sample submits a Fast model job and polls until it finishes:

import time
from google import genai
from google.genai import types

client = genai.Client()  # reads the credential from your environment

operation = client.models.generate_videos(
    model="veo-3.1-fast-generate-preview",
    prompt="Slow dolly shot through a rainy night market, steam rising from food stalls",
    config=types.GenerateVideosConfig(aspect_ratio="16:9", resolution="720p"),
)

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

video = operation.response.generated_videos[0]
client.files.download(file=video.video)
video.video.save("market.mp4")

Copy Prompt

Teams already running on Google Cloud can also reach Veo 3.1 through Vertex AI, where quota is handled at the project level and increases go through a quota request.

Picasso IA Without Code

Picasso IA runs the whole Veo 3.1 family from browser pages, so there is no Google Cloud project, billing account or quota dashboard to manage. Open Veo 3.1, Veo 3.1 Fast or Veo 3.1 Lite, write a prompt, and download the result.

Developers who want an API there get a different set of models. The Picasso IA API lives at https://api.picassoia.com/v1, uses a bearer token that starts with pia_sk_, and follows the Replicate pattern: create a prediction, poll it, fetch the output. It exposes four models, two of them for video: Picasso IA Video and Seedance 2.5 Lite. The Veo family is not on that list. Limits are 5 concurrent predictions per account, 4,000 character prompts and a 3 hour timeout. The site describes API predictions as currently free, but its docs also mention an Infinite plan, so check the pricing page before you build on that.

How to Use Veo 3.1 on PicassoIA

The workflow takes five minutes:

1. Open the Veo 3.1 page.
2. Write a prompt.
3. Choose duration, resolution and aspect ratio.
4. Add a start image, an end image or references if you need them.
5. Generate, review the clip and download it.

Write the Prompt

Veo 3.1 generates sound together with the picture, so write for both. A reliable structure is subject, action, camera move, lighting, then audio. For example: A street musician plays violin under a stone archway at dusk, slow push-in on his hands, warm lamplight, soft crowd murmur and echoing strings.

Use the negative prompt field to exclude things you do not want, such as text overlays or extra people.

Not sure how to phrase a scene? Ask an LLM to draft five variants first. Gemini 3.5 Flash is fast for brainstorming, and Claude Sonnet 5 or GPT 5.6 Terra work just as well. Paste your idea, ask for subject, action, camera, light and sound in one paragraph, then pick the best draft.

Pick Duration and Resolution

The Picasso IA form exposes these settings:

Use 9:16 for vertical social clips and 16:9 for everything else. For test runs, drop to 4 seconds at 720p, then switch to 8 seconds at 1080p once the prompt works. Reuse the same seed when you rerun a prompt so you compare changes fairly.

Add References or Frames

Three optional inputs give you more control:

• Start image: upload a photo to animate. Ideal sizes are 1280×720 or 720×1280, matching your aspect ratio.
• Last frame: add an end image together with the start image and Veo builds the transition between them.
• Reference images: upload 1 to 3 pictures to keep a subject consistent. This works only with 16:9 and 8 seconds, and the last frame is ignored when references are present.

Habits That Keep Costs Low

Draft Cheap, Render Once

Every extra render costs real money on a paid API, so treat the cheap models as your sketchbook. Run first drafts on Veo 3.1 Lite or Veo 3.1 Fast, and save Veo 3.1 for the final pass.

Other models are worth a test for rough drafts too. Veo 3 Fast makes videos with audio from text, LTX 2.5 Fast is listed for quick 4K output, and Seedance 2.5 Lite is listed as a free, unlimited generator with clips up to 10 seconds.

Download and Store Early

Google deletes generated files after 2 days, and one forgotten clip means paying again. Make saving part of your script, not a manual step. Keep a simple log with the prompt, model code, seed, resolution and file name for every render. When a client asks for "the version from last Tuesday," you will find it in seconds and avoid a repeat charge.

Try Veo 3.1 on Picasso IA

Reading about per second pricing is one thing. Watching your own prompt turn into a clip is another. Open Veo 3.1 on Picasso IA, type a scene, pick 720p and 4 seconds, and see how the model handles your idea before you spend a cent on API calls. Try a start image, add reference photos, swap in Veo 3.1 Fast for speed, and compare the results side by side. The more prompts you test now, the fewer renders you pay for later.