Google is widening access to Veo, its AI model for generating short video clips from images and prompts, by bringing it to Google Cloud customers through Vertex AI in private preview.
The announcement puts Veo in front of enterprise users after a slower rollout across Google Labs and YouTube Shorts. It also places the model in a market where OpenAI, Adobe, Runway, Luma, Meta and others are competing to define how AI-generated video will be used by creators, brands and developers.
What Google Is Making Available
Veo was unveiled in April as a video generation model that can produce 1080p clips of animals, objects and people up to six seconds long at either 24 or 30 frames per second. Google says the model can handle different visual and cinematic styles, including landscapes and time lapses, and can also edit footage it has already generated.
For Vertex AI customers, Google is starting with a private preview rather than broad general availability. Warren Barkley, senior director of product management at Google Cloud, described the wait as a matter of enterprise readiness, saying the company has augmented, hardened and improved Veo for customers on Vertex AI.
In its current Vertex AI preview form, Veo can create high definition videos in 720p, with support for 16:9 landscape and 9:16 portrait aspect ratios. Google says it plans to keep improving Veo in the same way it has improved other models such as Gemini on Vertex AI.
Early Enterprise Uses
Google named two customers tied to the launch. Quora is expected to bring Veo to Poe, its chatbot platform. Mondelez International, the owner of Oreo, is expected to use Veo with agency partners to create marketing content.
Those examples show the two sides of Google’s pitch. One is distribution through AI platforms where users can try different generative models. The other is commercial content production, where companies want faster ways to test and iterate on video ideas.
Google says Veo can interpret prompts involving VFX, including prompts such as enormous explosion. The company also says the model has some understanding of physics, including fluid dynamics, and supports masked editing for changing specific areas of a video. It is also technically capable of connecting footage into longer projects.
Where Veo Still Falls Short
Google’s broader claim is that Veo is competitive with leading video-generating models. But the source material also makes clear that the technology has familiar weaknesses.
Objects in Veo clips can disappear and return without clear consistency. The model can also make physics errors. One example given is cars that impossibly reverse on a dime.
These limitations matter because video is especially sensitive to continuity. A still image can hide many errors in a single frame. Video has to preserve objects, motion and spatial logic over time, which makes mistakes more visible and harder to ignore.
Training Data, Copyright and Safety Questions
Like other generative AI systems, Veo was trained on large amounts of footage. Google has not said exactly where it sources the data used to train its generative models. Asked specifically about Veo, Barkley said the model may be trained on some YouTube content in accordance with Google’s agreement with YouTube creators.
He also said Veo was trained on a variety of high-quality, video-description datasets that are heavily curated for safety and security, and that Google’s foundational models are trained primarily on publicly available sources.
The source article notes reporting by The New York Times in April that Google broadened its terms of service last year in part to allow more data to be used for AI training. Under the older terms, it was unclear whether YouTube data could be used for products beyond YouTube. The newer terms loosened that restriction.
Google offers tools that let webmasters block its bots from scraping training data from websites. It does not offer a way for creators to remove their works from existing training sets. Google says training models on publicly available data is fair use, while also saying it does not use customer data to train its models.
Another risk is regurgitation, where a model produces a near copy of training data. The source article notes that tools such as Runway’s have been found to generate stills substantially similar to copyrighted videos, creating possible legal risk for users.
Google’s answer for Veo includes prompt-level filters for violent and explicit content. The company also says its indemnity policy can defend eligible Veo users against copyright infringement claims. Barkley said Google plans to indemnify Veo outputs on Vertex AI when the model becomes generally available.
Slow Rollout, Bigger Stakes
Google has been adding Veo to more products gradually. In May, it brought Veo to Google Labs for select testers. In September, it announced a Veo integration for YouTube Shorts so creators could generate backgrounds and six-second video clips.
The company is also using SynthID, its proprietary watermarking technology, to place invisible markers in frames generated by Veo. The source article notes that SynthID is not foolproof against edits and that Google has not made the content ID piece available to third parties.
The business challenge is not only technical. Google has ceded some partnership ground to rivals that have moved quickly with producers, studios and creative agencies. Runway signed a deal with Lionsgate to train a custom model on the studio’s movie catalog, while OpenAI worked with brands and independent directors to show what Sora can do.
Google has said before that it was exploring Veo applications with artists including Donald Glover, also known as Childish Gambino. It gave no update on those outreach efforts in the announcement.
The tension is clear: Google is presenting Veo as a way to reduce costs and quickly iterate on video content, but that same pitch can worry creative workers. A 2024 study commissioned by the Animation Guild estimated that more than 100,000 U.S.-based film, television and animation jobs will be disrupted by AI by 2026.
Google has not given an ETA for Veo’s general availability in Vertex AI. Barkley also did not say when Veo might come to more Google platforms and services. The company says preview releases let it collect real-world feedback from selected enterprise customers before wider use.
Alongside the Veo news, Google also said Imagen 3, its flagship image generator, is now available to all Vertex AI customers without a waitlist. New customization and image editing features for Imagen 3 are available behind a separate waitlist.