Google brings Veo and Imagen 3 to enterprise cloud users

While everyone continues to watch OpenAI for signs of a public Sora release, Google just quietly slipped its flagship video model into the enterprise back door.

On Tuesday, Google announced that Veo, its AI model capable of generating short video clips from text and images, is now available in private preview on Vertex AI. Google Cloud claims this makes it the first hyperscaler to offer a still-image-to-video service natively. Alongside the preview, Google also opened its Imagen 3 image generator to all Vertex AI customers without a waitlist, though advanced customization and inpainting tools remain waitlisted.

Dialed down for enterprise

When Google unveiled Veo back in April, it teased 1080p clips up to six seconds long at 24 or 30 frames per second. It demonstrated a decent grasp of cinematic prompts, landscape timelapses, and even visual effects physics like fluid dynamics.

Now that it is actually hitting enterprise hands, the specs look a bit more practical. Warren Barkley, senior director of product management at Google Cloud, explained that the team spent months "hardening" the model for business use. On Vertex AI today, Veo generates 720p clips in either 16:9 landscape or 9:16 portrait. It also supports masked editing to tweak specific regions of a clip, along with timeline stitching tools.

Early enterprise testers are already moving in. Quora plans to bring Veo to its Poe platform to expand multi-modal access, while Oreo parent company Mondelez International is testing it with agency partners to generate marketing assets.

Physics glitches and copyright shields

Veo may compete head-to-head with Runway, Luma, Adobe, and Sora, but it still suffers from classic generative AI weirdness. Objects occasionally pop in and out of thin air, and cars can still inexplicably reverse on a dime without respecting inertia.

Then there is the training question. Barkley acknowledged that Veo "may" be trained on "some" YouTube content under creator agreements, leaning mostly on public sources. While Google offers web scrapers an opt-out bot, human creators still have no way to pull their work out of existing datasets.

To soothe corporate legal teams worried about regurgitation and copyright lawsuits, Google announced plans to extend its copyright indemnification policy to Veo outputs once the model reaches general availability. Video generations will also feature Google's invisible SynthID watermark, though third parties still cannot access detection tools to verify them independently.

Why it matters

This move highlights a massive split in the AI video landscape: startups are chasing Hollywood, while tech giants are turning video into mundane cloud infrastructure.

For businesses, accessing video generation directly through a standard cloud provider simplifies billing, security compliance, and deployment. Integrations like Quora's Poe demonstrate that developers will soon treat Veo as just another API endpoint. However, at 720p and capped at six seconds with noticeable physics hallucinations, teams should view it as a rapid prototyping tool for storyboards and social drafts, not finished production pipelines.

We think Google's enterprise-first approach is strategically sound, if visibly cautious. Competitors like Runway have rushed to strike studio deals with Lionsgate, risking friction with entertainment unions who fear industry layoffs. Google, by contrast, is playing the boring game: slipping Veo into marketing workflows and YouTube Shorts backgrounds. Tucking the model behind an enterprise preview gives Google time to iron out consistency flaws, though relying on vaguely licensed YouTube data means legal friction will inevitably linger.