Veo is Google DeepMind's AI model for generating video. Version 3.1 produces clips of 4, 6 or 8 seconds with sound already in place: speech, ambient noise, music. You can use it in Flow, in Google's enterprise products and through gateways such as codexkey. Year-old instructions are out of date: on 22 October 2026 Google shut down the Veo 3.1 preview models in the Gemini API.
What Veo 3.1 can do
Sound arrived in the lineup with Veo 3: on the Veo page of the DeepMind site, Google describes it as native generation of effects, ambient noise and dialogue together with the picture. Version 3.1 came out on 15 October 2025 in public preview in the Gemini API and added:
- reference images, up to three per clip;
- a clip between a first and a last frame;
- extension of a scene that Veo created;
- durations of 4, 6 and 8 seconds.
The Veo page also lists ways of working with finished footage: add an object to a scene, remove an unwanted one, set the path of a moving item, carry a style over from a reference image. Which of these is available in your interface depends on the product and the region, so read the list as a description of what the model can do, not as a promise for every service.
Later, on 31 March 2026, a lighter Veo 3.1 Lite appeared for quick iterations (Gemini API changelog). Google marks the output with a SynthID watermark. The company also says it blocks harmful requests and results, so the model may refuse some scenes however many times you rewrite the prompt. The developer itself warns that natural speech in short phrases is still being refined, and incoherent speech is possible.
What changed by 29 October 2026
This is the main news for anyone searching for "how to use Veo 3". The picture right now:
| Where | What is there on 29 October 2026 |
|---|---|
| Gemini app | video is made by Gemini Omni, not Veo, per the Gemini Apps help; Russia is not on the app's country list |
| Gemini API | the Veo 3.1 previews (generate, fast, lite) were shut down on 22 October 2026; the replacement is gemini-omni-1.1-flash, which reached GA on 27 August 2026 |
| Flow | Veo 3.1 Lite, Fast and Quality, plus Gemini Omni Flash, for credits |
| Gemini Enterprise Agent Platform | Google names the GA models there as the replacement for Gemini API users |
| codexkey | veo-3.1 is listed in the catalog; check the model page for status |
On the Gemini API there is a single source: the model shutdowns page names gemini-omni-1.1-flash as the recommended replacement. The new model supports video extension, a clip between two images, and a choice of 360p, 720p (the default), 1080p and 4K, with 1080p and 4K produced by upscaling.
How Gemini Omni Flash differs from Veo 3.1 shows in Google's description and in the Flow credit table:
| Parameter | Veo 3.1 | Gemini Omni Flash |
|---|---|---|
| Durations in Flow | 4, 6, 8 seconds | 4, 6, 8, 10 seconds |
| What it accepts as input | text, an image, up to three references, a first and a last frame | text, images, audio and video at once |
| Editing a finished clip | scene extension | conversational editing over several turns |
| Resolution in the Gemini API | not checked | 360p, 720p, 1080p, 4K (the last two by upscaling) |
What follows for the reader? If you write code for Google, look at Gemini Omni Flash rather than Veo 3.1. If you specifically need Veo, Flow and gateways remain.
Veo 3.1 for free: what Flow gives
Flow hands out 50 credits a day regardless of subscription, and the leftover does not carry over to the next day (Flow credits help). What one generation costs for users without an Ultra subscription:
| Model | Credits per generation | How many fit in 50 credits |
|---|---|---|
| Veo 3.1 Lite | 10 | 5 generations |
| Veo 3.1 Fast | 20 | 2 generations |
| Veo 3.1 Quality (8 seconds) | 100 | none |
| Gemini Omni Flash, 720p, 8 seconds | 12 | 4 generations |
The last column is our own calculation, not a Google statement. It is enough for tests and not enough for a project. For Russia this route has one problem: the country is missing from the list where Flow is available (Flow list), and the help page says outright that a VPN will not open access. Russia is also absent from the list of regions for the Gemini API and Google AI Studio (region list).
Veo 3 subscription: what Google AI plans give
Google AI subscriptions add monthly credits to Flow's 50 daily ones: Plus 200, Pro 1,000, Ultra at $100 10,000, Ultra at $200 25,000. On Pro that is 10 Quality generations or 50 Fast ones a month at the prices in the table above. Subscription credits expire at the end of the billing period. Flow is not supported in Russia, so we do not describe Google's payment methods.
Which Veo 3.1 tier to pick: Lite, Fast or Quality
Flow shows the price in credits directly: Lite costs 10, Fast 20, Quality 100, and Quality is available only for 8-second clips. The prices differ tenfold, and that sets the order of work. Run draft versions of a scene on Lite or Fast and do the final pass on Quality, so the expensive tier is not spent on a clip you will have to reshoot because of a prompt mistake. In the codexkey catalog the same three tiers are named lite, fast and quality.
How to write a prompt for Veo 3.1
In Google's own examples the sound is described in a separate sentence, and the duration is stated in the text. Use the same scheme: shot, action, camera, sound. Our example:
"Close-up: an elderly watchmaker raises a loupe to his eye, snow falling outside the workshop window. The camera slowly pushes in. Over 8 seconds he spots a faulty gear and says quietly: 'There you are'. Sound: the ticking of dozens of clocks, the muffled tap of snow on the glass, no music."
Mind the length of spoken lines: a person speaks roughly two to three words a second, so 4 seconds fit a line of about ten words, and 8 seconds fit a couple of short lines. If a line is longer than the clip, the model either cuts it off or smears the speech. This is our practical estimate, not a parameter from the documentation.
Whether a Russian prompt works as well as an English one, we did not find in the documentation. Compare both versions on one scene before spending your limit on long ones.
Veo 3.1 through the codexkey API
In codexkey a video job is created with POST /v1/videos, the status is read with GET /v1/videos/{id}, and the file is fetched with GET /v1/videos/{id}/content. The format is compatible with the OpenAI Videos API (reference). You create a key by following the quickstart.
curl https://api.codexkey.online/v1/videos \
-H "Authorization: Bearer sk_cdx_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "veo-3.1", "prompt": "A paper boat sails down a rainy street, with raindrops and distant thunder"}'
curl https://api.codexkey.online/v1/videos/VIDEO_ID \
-H "Authorization: Bearer sk_cdx_YOUR_KEY"
curl https://api.codexkey.online/v1/videos/VIDEO_ID/content \
-H "Authorization: Bearer sk_cdx_YOUR_KEY" \
-o video.mp4
In the codexkey catalog veo-3.1 lists the tiers lite, fast and quality, durations of 4, 6 and 8 seconds, formats 16:9 and 9:16, and the modes "from text", "first and last frames" and "from references". Take exact field names and allowed values from the model page in the catalog: they can change.
Before you launch, run through a short checklist:
- Open the
veo-3.1page in the catalog and see whether it carries an unavailable mark. - For a first try, pick the lite tier and a 4-second clip.
- Describe the sound in the prompt as a separate sentence, or you will be guessing at what came out.
- Download the result right after the completed status and keep it on your side.
- Compare the result with another model on the same scene, as in the overview of AI video generators.
One caveat. The shutdown of the preview models concerns the Gemini API at Google itself. Whether veo-3.1 works for you today, check with a trial request on the shortest and cheapest tier. There is no public price list for video, the price is shown on the model page, and money is charged only for a successful result. Payment is in rubles: a Russian bank card, SBP (the Russian Faster Payments System) or cryptocurrency, no VPN needed.
Other video models under one key, from Seedance to Kling, are covered in the articles on Seedance 2.5 and on Kling AI. If you need Google's text models, read about the Gemini API from Russia.
FAQ
Veo 3 for free: can you try it without paying?
Flow gives 50 free credits a day: that is five Veo 3.1 Lite generations or two Fast ones. Flow is not supported in Russia, so the realistic route for us is the API, paid per successful result.
How does Veo 3.1 differ from Veo 3?
Veo 3 brought native sound. Version 3.1 added reference images (up to three), a clip between a first and a last frame, scene extension, and durations of 4, 6 and 8 seconds.
Does Veo 3 work in Russia?
Google does not support direct access from Russia: the country is missing from the lists for the Gemini API, the Gemini app and Flow, and a VPN, as the Flow help says, does not open access. Through the codexkey gateway the veo-3.1 model is called over the API without a VPN.
Which Veo 3.1 tier should I choose: Lite, Fast or Quality?
Lite and Fast suit drafts and testing an idea, Quality suits the final. In Flow the prices differ tenfold: Lite costs 10 credits, Quality 100, and Quality works only with 8-second clips.
What is Veo 3 Flow?
Flow is Google's tool for video and scenes, built on the Veo and Gemini Omni models. It spends credits: 50 a day free and more with a Google AI subscription.
Veo 3 subscription: how many credits do the plans give?
On top of the 50 daily credits, per month: Google AI Plus gives 200, Pro 1,000, Ultra at $100 10,000, Ultra at $200 25,000. Check subscription prices and terms on the Google One page for your region.
What happened to Veo 3.1 in the Gemini API?
The preview models veo-3.1-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-lite-generate-preview were shut down on 22 October 2026. Google recommends Gemini Omni Flash, gemini-omni-1.1-flash. The stable veo-3.0-generate-001, veo-3.0-fast-generate-001 and veo-2.0-generate-001 were shut down by Google back on 30 June 2026, so old code samples using them no longer work either.