A practical Google Veo guide with verified API pricing, model differences, native audio, and lessons from François Aubry's published tests.
Google Veo is Google DeepMind's video-generation family, built to turn prompts and images into short clips with native audio. Its appeal is easy to understand: a scene can arrive with movement, dialogue, and atmosphere already connected. Getting those elements to remain coherent across several shots is the harder part.
This guide covers Veo 3.1, current API prices, and practical lessons from a named creator's published experience.
Veo generates video from text or image inputs. Native audio can include speech, effects, and ambient sound. It is useful for ad concepts, product animations, previsualization, and short illustrative scenes.
Google's showcase demonstrates the possibilities, but a polished example does not tell you how many attempts a similar brief will require.
The current Veo family includes standard, Fast, and Lite variants. Google's March 31, 2026 announcement introduced Lite as a lower-cost developer option.
| Variant | Practical starting point |
|---|---|
| Veo 3.1 standard | Quality-focused generation; test against your brief |
| Veo 3.1 Fast | Faster iteration and lower API cost |
| Veo 3.1 Lite | Budget-conscious generation at 720p or 1080p |
Lite supports text-to-video, image-to-video, landscape and portrait formats, and durations of 4, 6, or 8 seconds. It does not support 4K output. Avoid assuming every variant has identical reference, extension, or resolution options.
Google's current video API overview recommends Gemini Omni Flash as the default for video generation and conversational editing. It points to Veo 3.1 when capabilities such as scene extension, last-frame control, or compatibility with existing pipelines matter.
For a new application, compare those workflows before choosing a model. A guide that describes Veo as the only current Google video option would miss an important decision.
According to the Veo developer guide, Veo 3.1 supports native audio, landscape and portrait output, first-and-last-frame generation, extension of Veo-generated videos, and direction using up to three reference images.
These controls serve different jobs. A starting frame anchors composition; references guide subjects; an ending frame sets a destination; extension continues an existing clip. They do not guarantee flawless transitions or character continuity. Check the exact variant's restrictions before combining features.
Engineer and founder François Aubry documented his tests in a DataCamp tutorial published October 22, 2025.
He made a story about a flame and a water droplet, reusing frames as starting points for later clips. The result required external editing because the clips did not join cleanly. The flame's voice changed, and the soundtrack lacked continuity.
In another test, he combined a person, outfit, and location. Details from an outfit reference's background leaked into the generated setting, and hand gestures were unconvincing. His practical advice was to use neutral backgrounds for reference assets.
He also found the API smoother than Flow, while reporting interface frustration and failed generations at that time. Those are dated observations, not a claim that today's interface behaves identically.
The lesson for a working creator: plan audio continuity as carefully as visual continuity, and clean references before blaming the prompt. A beautiful frame can still create an awkward edit.
Start from Google's official links and confirm which model your account actually exposes:
Availability and allowances vary. A consumer subscription is not the same billing arrangement as API usage.
The official Gemini API pricing page, checked October 8, 2026, lists these video-with-audio rates in USD:
| Variant | 720p / second | 1080p / second | 4K / second |
|---|---|---|---|
| Standard | $0.40 | $0.40 | $0.60 |
| Fast | $0.10 | $0.12 | $0.30 |
| Lite | $0.05 | $0.08 | Not supported |
An eight-second 720p clip therefore costs $3.20 standard, $0.80 Fast, or $0.40 Lite at these rates. Four successfully generated Fast attempts cost $3.20, even if only one makes the edit.
The page lists no free API tier for these models and says you are charged only when the video is successfully generated. Consumer credits and Vertex AI rates should be checked separately.
Prompt formula: Subject + Action + Setting + Camera + Lighting + Audio
> A barista pours latte art into a ceramic cup in a sunlit café. Slow close-up push-in, warm morning light, gentle steam. Quiet café chatter and a soft espresso-machine hiss. One continuous shot.
Keep dialogue short, identify the speaker, and put exact words in quotation marks. Change one variable between attempts.
Review generated speech, transitions, and visual details before delivery. Google's DeepMind page acknowledges continuing work on natural, consistent spoken audio.
Confirm terms for your access route and obtain rights to uploaded assets and likenesses. Google marks Veo outputs with SynthID; follow any additional disclosure requirements where you publish.
As checked October 8, 2026, the official DeepMind Veo page identifies Veo 3.1. The official sources reviewed for this guide did not establish a Veo 4 release. Also, 4K is already listed for supported Veo 3.1 variants, so a 4K claim alone is no evidence of Veo 4.
Veo is worth testing when generated audio, references, or scene extension fit a real project. Start with a short sequence, track every usable and rejected attempt, and listen across cuts. The cost of a finished scene—and the work needed to make it coherent—is a better measure than the first impressive clip.
Google Veo is Google DeepMind's video-generation family. It creates short clips from text or images, with native audio in the Veo 3.1 family.
Google lists standard, Fast, and Lite variants. Their prices and supported features differ, so check the exact model and access route.
No. Veo 3.1 Lite supports four-, six-, and eight-second generations. Duration restrictions vary by variant and feature.
Yes. Veo 3.1 generates native audio, including dialogue, effects, and ambience. Review speech and continuity before publishing.
At the Gemini API rates checked October 8, 2026, an eight-second 720p video with audio costs $3.20 on standard, $0.80 on Fast, or $0.40 on Lite. Successfully generated retakes add to the total.
The official pricing page lists no free API tier for Veo 3.1 standard, Fast, or Lite. Consumer allowances and subscriptions are separate.
Google recommends Gemini Omni Flash as the default for video generation and conversational editing. It points to Veo 3.1 for capabilities such as scene extension, last-frame control, and existing pipeline compatibility.
The official sources reviewed on October 8, 2026 did not establish a Veo 4 release; the DeepMind Veo page identifies Veo 3.1. Check Google's official announcements for updates.
