Create video with Google Veo 3.1

What Google Veo 3.1 is

Veo 3.1 is Google DeepMind's video generation model, announced on 15 October 2025 as the successor to Veo 3. Google's launch post describes it in one sentence: richer audio, more narrative control, and enhanced realism that captures true-to-life textures, with stronger prompt adherence and improved audiovisual quality when turning images into videos.

The launch post names the capability changes directly. Audio support reached existing features for the first time, so Ingredients to Video, Frames to Video and Extend all produce sound. Scene extension got its own section, and the post describes more granular control over editing, including inserting, removing or changing elements of a clip.

A second official post on 13 January 2026 continued the line: richer dialogue and storytelling from Ingredients to Video even with short prompts, identity consistency that keeps a character looking the same across scenes, native vertical outputs for portrait formats, and state-of-the-art upscaling to 1080p and 4K. That update rolled out across the Gemini app, YouTube, Flow, Google Vids, the Gemini API and Vertex AI.

The three variants Google publishes

Google's Vertex AI model card, last updated 10 October 2026, documents three Veo 3.1 variants with separate model IDs and capability sets. The rows below quote that card.

Every variant on the card carries Content Credentials (C2PA), so clips produced by the API ship with provenance metadata. The card also marks each variant's serving region as us-central1 and the prompt language as English.

What the model card specifies

These are the shared technical specifications from the same Vertex AI card, quoted as written.

Two rows deserve a plain reading. The 4K output row belongs to the Generate variant only, and the English-only row means the prompt text should be written in English regardless of the language you want spoken in the clip.

The two benchmarks Google publishes

The DeepMind model page publishes two preference benchmarks for Veo 3.1, both run on MovieGenBench, a dataset Meta released. The charts are reproduced in the gallery below.

Both are Google's own evaluations of its own model, published on the model's marketing page. They are first-party numbers, not independent audits, and this page repeats them as exactly that.

Prompts Google publishes for this model

Every prompt below is published in the Veo section of Google's Gemini API documentation and reproduced word for word. The code block beside each example names the model veo-3.1-generate-preview, so these strings are bound to the 3.1 line by Google's own docs.

The developer docs publish one more string worth noting beside its own video: a short prompt for extending an existing clip (track the butterfly into the garden as it lands on an orange origami flower), which is how the extend feature is demonstrated. The demonstration clips on this page are different: Google's launch videos have no published per-clip prompts, so none are shown on their cards.

A close up of two people staring at a cryptic drawing on a wall, torchlight flickering. A man murmurs, 'This must be it. That's the secret code.' The woman looks at him and whispering excitedly, 'What did you find?'

Drone shot following a classic red convertible driven by a man along a winding coastal road at sunset, waves crashing against the rocks below. The convertible accelerates fast and the engine roars loudly.

A whimsical stop-motion animation of a tiny robot tending to a garden of glowing mushrooms on a miniature planet.

The video opens with a medium, eye-level shot of a beautiful woman with dark hair and warm brown eyes. She wears a magnificent, high-fashion flamingo dress with layers of pink and fuchsia feathers, complemented by whimsical pink, heart-shaped sunglasses. She walks with serene confidence through the crystal-clear, shallow turquoise water of a sun-drenched lagoon. The camera slowly pulls back to a medium-wide shot, revealing the breathtaking scene as the dress's long train glides and floats gracefully on the water's surface behind her. The cinematic, dreamlike atmosphere is enhanced by the vibrant colors of the dress against the serene, minimalist landscape, capturing a moment of pure elegance and high-fashion fantasy.

What this workspace exposes

The picker is the authority on what you can submit here, and it is narrower than the model card. The rows below are read from this site's own model configuration and the public product catalog.

The route row matters and is easy to miss: clips generated here come from the Fast variant, so the 4K option in the picker is a request parameter, not evidence that a given clip was rendered by the full Generate model. The audio row runs the same way; generated audio is applied by default on the service side and this picker offers no switch for it.

What this workspace does not expose

The model card documents capabilities this picker has no control for. They are listed so you do not look for them.

• First and last frame mode: documented on the card for Generate and Fast, not offered here.
• Extend: the documented way to continue an existing clip is not offered here.
• Reference-image generation beyond one image: Google's developer docs describe up to 3 reference images; this workspace accepts one in image-to-video.
• Variant choice: there is no picker between Generate, Fast and Lite. The route is fixed to Fast.
• Batch generation: the card allows up to 4 outputs per prompt; this workspace submits one task at a time.
• An audio switch: generated audio is default-on service-side, with no toggle in the request.

How to generate here

Four steps, in the order the workspace asks for them.

1. Pick the mode: text to video, or image to video when a still should drive the shot.
2. Write the prompt in English, or open one from the list above and copy it.
3. Set duration between 4 and 8 seconds, resolution and aspect ratio, then attach one reference image if you are animating a still.
4. Read the quote, sign in when the workspace asks, then submit. If a task fails, read its status before retrying so you do not submit a duplicate.

Why other sites quote different numbers

Pages ranking for this model quote prices and capabilities in incompatible units. None of them is the number that applies to a task submitted here.

The prices below were read from the competitor pages themselves in October 2026. They are third-party figures for third-party services, they change, and their units do not convert cleanly into each other.

The only figure that binds a task on this page is the quote the workspace prints immediately before submission, which is resolved server-side from the live catalog.

• convly.ai, a model-directory page, states pricing from 0.05 US dollars per second of video.
• YeLin AI, an API reseller, states pricing from 0.15 US dollars per generation across 8 model variants, in its own OpenAI-compatible interface.
• veo3ai.io, a tool site, states 10 credits per 8-second clip with a signup allowance, in Chinese.
• Fastlane, a marketing blog, describes 48 kHz native audio and Ingredients to Video for product shots; it sells no generation itself.

What this page does not promise

A model page should say where its own claims stop. These are the boundaries for this one.

• No free allowance is promised. Whatever credits exist are shown in your account, not here.
• The four clips in the gallery are Google DeepMind's official launch demonstrations. None was generated on this site.
• The two benchmark figures are Google's own evaluations of its own model. This page ran no comparison of its own.
• The launch claims (richer audio, stronger prompt adherence, identity consistency, upscaling) are Google's descriptions of its own product, repeated as such.
• The official launch videos carry no published prompts, so their gallery cards show no prompt rather than an invented one.
• Third-party prices are the numbers those businesses published for their own services in October 2026; they are not quotes for tasks submitted here.
• Latency, queue behaviour and final billing depend on the service at the moment you submit, not on anything stated here.

Where the samples and numbers come from

Every figure and every clip on this page traces back to one of the sources below. Media licensing is stated with the gallery cards.