Updated: September 3, 2026 · Docs · Clothoff AI Editorial Team
What is Veo?
Google DeepMind's video model: API-only, every output carries an invisible SynthID watermark, and Google's policies block sexual content and non-consensual likeness.
Veo is a closed video generation model developed by Google DeepMind and served only through Google APIs, with Veo 3 generally available on Vertex AI from Jul 29, 2025, Veo 3.1 released in 2025 and Veo 3.1 Lite added on Apr 3, 2026. It generates text-to-video and image-to-video clips with audio, and every frame carries an invisible SynthID watermark.
| Version | Date | Developer | License / Access | Filters |
|---|---|---|---|---|
| Veo 3 (veo-3.0-generate-001) | Jul 29, 2025 (GA); discontinued Jun 30, 2026 | Google DeepMind / Google Cloud | API via Vertex AI and Gemini API | SynthID watermark; Google generative AI policies |
| Veo 3.1 (veo-3.1-generate-001) | 2025 (day not restated on the version table) | Google DeepMind / Google Cloud | API via Vertex AI and Gemini API; Gemini app | SynthID watermark; person-generation controls per API |
| Veo 3.1 Lite | Apr 3, 2026 | Google Cloud | API via Vertex AI, Media Studio | SynthID watermark; same policies |
Google publishes version status on the Vertex AI model-versions page; older Veo 2 endpoints have also been retired and are omitted here.
Which Veo versions are current?
Google Cloud's Vertex AI version table lists veo-3.0-generate-001 with general availability on Jul 29, 2025 and a discontinuation date of Jun 30, 2026, which moves production traffic to Veo 3.1. Veo 3.1 adds reference images, first-and-last-frame control and scene extension.
Veo 3.1 Lite, announced Apr 3, 2026, is the lower-cost variant for high-volume use, available in Vertex AI Media Studio and through the API. There is no downloadable checkpoint for any version; Veo runs only on Google infrastructure.
Does Veo allow NSFW output?
No. Veo is governed by Google's Generative AI Prohibited Use Policy and the Gemini API terms, which exclude sexually explicit content and depictions of real people without consent. Requests are filtered before and after generation, and the API exposes person-generation settings that further limit outputs containing people.
The stronger safeguard is provenance: every Veo frame carries a SynthID watermark that survives cropping and compression and can be checked with Google's detector. That means a Veo clip can be traced as synthetic even when the visible label is removed, unlike output from unwatermarked open models.
“SynthID embeds digital watermarks directly into AI-generated images, audio, text or video.”
— Google DeepMind, SynthID, accessed Sep 3, 2026
How is Veo used in tools we review?
None of the tools in our rankings claims a Veo backbone, and Google's policy would not permit the outputs those tools produce. We include Veo because it is the clearest example of a watermarked, moderated video model against which readers can judge the scoring axes we apply to third-party services.
When a reviewed tool offers video, as noted in the Pornworks review, the practical questions are retention, deletion and whether outputs carry any provenance mark. Veo answers the last question by design; most services we tested do not.
What does SynthID mean for provenance?
SynthID is an invisible watermark that Google applies to images, audio, text and video generated by its models. It is not C2PA metadata, which lives in the file container and is stripped by many platforms; SynthID is embedded in the pixels and audio samples themselves.
For anyone assessing a clip, a positive SynthID detection is strong evidence of Google-model origin, while a negative result proves nothing about other models. See SynthID and watermarking for the detection workflow and its limits.
Consent note: tools in our rankings are documented for use only with photos of yourself or consenting adults 18+. See Responsible AI; if an image of you was misused, report NCII.
Sources
- Google Cloud — Veo model versions on Vertex AI — GA Jul 29, 2025; discontinuation Jun 30, 2026, accessed Sep 3, 2026.
- Google Cloud blog — Veo 3.1 Lite — announced Apr 3, 2026, accessed Sep 3, 2026.
- Google AI for Developers — Veo video generation — Gemini API access and parameters, accessed Sep 3, 2026.
- Google DeepMind — SynthID — watermarking of images, audio, text and video, accessed Sep 3, 2026.
- Google DeepMind — Veo model page — model overview, accessed Sep 3, 2026.
Reference entry by the Clothoff AI Editorial Team; not legal advice. The access date marks the last verification.
Frequently asked questions about Veo
Is Veo open source?
No. Veo is a closed model developed by Google DeepMind and served only through Vertex AI, the Gemini API and Google consumer apps. There are no published weights, and every generation is processed on Google servers under the Gemini API terms and Google's generative AI usage policies.
When were Veo 3 and Veo 3.1 released?
Veo 3 reached general availability on Vertex AI on Jul 29, 2025 and is scheduled for discontinuation on Jun 30, 2026 according to Google Cloud's version table. Veo 3.1 followed later in 2025 with reference images and frame control, and Veo 3.1 Lite was announced on Apr 3, 2026.
Does every Veo video carry a watermark?
Yes. Google states that Veo outputs are marked with SynthID, an invisible watermark embedded in the pixels and audio rather than in file metadata. It survives common edits such as cropping and compression and can be verified with Google's detection tools, which is why Veo clips remain traceable.
Can Veo generate sexual content or real people?
No. Google's Generative AI Prohibited Use Policy excludes sexually explicit content and non-consensual depictions of real people, and the API adds person-generation controls. Prompts and outputs are filtered, so Veo cannot serve the intimate-imagery workflows that unfiltered open checkpoints allow.
What is the difference between Veo 3.1 and Veo 3.1 Lite?
Veo 3.1 is the full model with reference images, first-and-last-frame control and scene extension. Veo 3.1 Lite, announced Apr 3, 2026, trades some quality for lower cost and is aimed at high-volume production through Vertex AI Media Studio and the API, while keeping SynthID marking and the same policies.
Do the tools we review run on Veo?
No reviewed tool names Veo, and Google's policies would block the explicit outputs those services produce. Our rankings therefore do not rely on backbone claims; they score tested behavior such as upload handling, retention, deletion, moderation and support, weighted per the published methodology.