What is Veo 3.1 (the model people call Veo 3), and how do you get access?

Veo is a family of AI video models from Google DeepMind. You describe a scene in words, or give it a still image to start from, and it produces a short clip with sound. Veo 3 introduced built-in audio, and the latest version Google lists on its Veo page is Veo 3.1. If you came looking for "Veo 3", Veo 3.1 is what you can actually use today.

This page covers what Veo 3.1 is, what it does well and where it is weak, the order I would try the access routes in, what Google's plan page showed on 4 November 2026, and a change that broke a lot of old tutorials. It is separate from Higgsfield vs Google Veo and Flow, which compares platforms. Here I keep to the model and the access map. Some of it runs through Higgsfield, a web platform that sells access to many AI image and video models from one login.

This page may contain affiliate links from which Blur Studio earns commissions.

Veo 3 or 3.1: which one you can use now

Google DeepMind's Veo page lists Veo 3.1 as its leading video model. Wikipedia lists 15 October 2025 as the Veo 3.1 release date. Older articles and videos say Veo 3, and many use it as a general name for the whole family. For a buyer, the practical rule is simple: if a guide names Veo 3, check whether the button you click in the product says 3.1, 3.1 Fast or 3.1 Lite. Fast is a cheaper, quicker tier and Lite is the lowest-cost one, per Google's and Higgsfield's lists.

What Veo does well

Per Google's own Veo page, Veo 3.1 offers:

• Built-in sound. Dialogue-style audio, ambient noise and effects come with the clip, so you do not add them in a separate step.
• High resolutions. Google lists 1080p and 4K outputs.
• Ingredients to video. You give it reference images (a person, an object, a setting) and it builds a clip with them.
• First and last frame. You give a starting picture and an ending picture and the model fills in the move.
• Scene extension. It continues a clip further.
• SynthID. Google marks its videos with SynthID, an invisible watermark meant to show that AI made them.

Treat these as Google's descriptions. Which of them you get depends on the product you use, since each route exposes a different slice.

Where it is weak

• Speech. Google's page says natural spoken audio is still in development, especially for shorter speech segments, and that the team is working to remove incoherent speech. Do not plan a spokesperson ad around it without a test.
• Clip length. Clips are short. On Higgsfield, Veo 3.1 runs 4, 6 or 8 seconds. An ad is joined from several shots.
• Exact products. As with any video model, text, logos and hands can drift. Filmed footage is safer for a product that must look exactly right.
• Not invisible. Because of SynthID, you cannot assume a client's viewers or a platform's detector will miss the AI origin. Plan the disclosure instead of hoping.

The access routes, in the order I would try them

Google's Veo page lists these ways in: the Gemini app, Flow, the Gemini API, Google AI Studio, Google Vids and the Gemini Enterprise Agent Platform. Plain meanings:

Route	What it is	Who it suits
Gemini app	Google's AI assistant app and website	First try, no setup
Flow	Google's AI film tool for building scenes from clips	People making a short story or ad from several shots
Google Vids	Google's video maker for work presentations	Internal videos in a Google Workspace company
Gemini API and Google AI Studio	Developer tools: AI Studio is a web workbench, the API lets software call the model	Developers
Gemini Enterprise Agent Platform	Google's business platform for building AI agents	Larger companies
Third-party platforms	Sites that resell access, such as Higgsfield	Teams that use many models

My ladder for a small business: try it in the Gemini app, move to Flow if you need to join shots, and use the API or a platform only if you need automation or many models. Do not buy a top plan to make one video.

What the plans look like (they depend on your country)

Google's AI plans page is regional. On 4 November 2026, the view available showed prices in United Arab Emirates dirhams (AED), and showed no US dollar prices. So no US prices are given here. What that view listed:

• Free: AED 0, with Flow access and limited use of the Nano Banana Pro image model.
• Plus: AED 18.99, with 200 Flow units.
• Pro: AED 76.99, with 1,000 Flow units (Flow's credits).
• Ultra: from AED 369.99 (the higher option is AED 769.99), with 10,000 or 25,000 Flow units depending on the option.

The page did not name Veo 3.1 on any plan card. It says Flow units are used with Gemini Omni Flash, Google's newer video model, so the plan page does not tell you which Veo versions each plan includes. Several news and comparison sites quote US prices for Ultra that disagree with each other, including a figure reported as a price cut. None is confirmed on a Google page, so they are left out. To see your number, open gemini.google/subscriptions from your own country and read the plan card. Availability of Veo itself also varies by country, so look at your region's page.

API users: the 22 October 2026 change

This is the trap for anyone following an older tutorial. Google's AI Studio video documentation says the Veo 3.1 Preview models in the Gemini API (the preview, Fast and Lite preview versions) were shut down on 22 October 2026. Any script or no-code automation that calls those model names will fail. Google's page recommends moving to Gemini Omni Flash (named gemini-omni-1.1-flash) or to the generally available models on the Gemini Enterprise Agent Platform. Gemini Omni Flash is a newer Google model that can also make video.

Google's AI Studio pricing page did not list Veo prices in the part that loaded. It showed Gemini Omni Flash video at $17.50 per million output tokens and a footnote of 5,792 tokens per second of 720p video, which works out to about $0.10 per second (17.50 x 5,792 / 1,000,000 = $0.1014). Treat per-second Veo prices on third-party pages as unconfirmed. If your agency or developer built a workflow on the old preview names, ask them to check it now.

Watermarks and disclosure

Besides SynthID, check the rules of the platform where the video will run. Many platforms ask you to label realistic AI video, and laws on AI in advertising are changing. This is information, not legal advice: check local law and the platform's current rules, and read the rights checklist for AI video in paid ads.

Using Veo 3.1 through Higgsfield

As of 4 November 2026, Higgsfield's model list shows Google Veo 3.1 (4, 6 or 8 seconds, a fast or preview variant, in 16:9 and 9:16, with an image as the start frame) and Veo 3.1 Lite (which also takes an end frame). Higgsfield's pricing table lists Veo 3.1 Fast at about 16 credits for a 4 second clip. The value is that you can put a Veo shot next to a Kling or Seedance shot for the same scene in one account and pay in one set of credits. Plans and what a month costs are in the pricing guide, and the platform overview is in What is Higgsfield.

Free accounts carry a Higgsfield watermark on every output. The pricing table marks commercial use as not included for Free, while the help center says the terms do not tie commercial use to a plan. I would use a paid plan for client work and read the terms first. For the platform comparison with Flow, see Higgsfield vs Google Veo and Flow.

A 6-shot ad test

Because clips are about 8 seconds, an ad is a set of shots. A plan you can run:

1. Write a six-shot storyboard for a 30 second ad: wide, product close-up, hands, reaction, detail, end card (the end card can be plain design, not AI).
2. Generate each shot on the cheap tier first (Lite or Fast).
3. Pick the best three and re-run them on the main tier.
4. Test any speech or sound line separately, since Google says speech is still in development.
5. Join the clips in an editor and watch the full ad with sound on.

Pass: product text and colours are right, the clips match each other in light and style, no audio glitch, and you can disclose the AI use in line with the platform rules. Fail: look changes from shot to shot, speech is off-sync, or the product shape changes. Record cost per approved shot, not per attempt. For more on planning a first AI ad, see your first AI video ad.

Who should skip Veo, and who should skip Higgsfield for it

• Skip Veo if you need clips longer than about 8 seconds in one go. Kling and Seedance are listed with longer limits on Higgsfield (Seedance 2.5 up to 30 seconds), and Dreamina's page shows 30-second Seedance 2.5 examples. See Higgsfield vs Kling and what Seedance is.
• Skip Veo for a speaking spokesperson. Filmed footage or a stock clip is safer for that.
• Skip Higgsfield for Veo if Veo is the only model you want and you already pay for a Google plan that includes it. Check what your plan covers on Google's page first.
• Skip the API route if you are not a developer. The Gemini app and Flow are simpler.

How I'd approach this

Say "Veo 3.1" in your notes and briefs so nobody hunts for the old name. Try one shot in the Gemini app before you pay for anything, then decide whether Flow, a platform or the API fits. Check the plan card for your own country, never a price from memory. If you have any automation that used the old API model names, fix it now. Plan ads as a set of shots, test sound separately and keep the cost per approved shot. If you want the process built once, with approvals and a shared asset library, Blur Studio builds that kind of workflow.