On Tuesday, Google introduced Veo, its AI-powered video generation model, announcing its availability in private preview for customers utilizing Vertex AI, Google Cloud’s platform for AI development. Veo enables users to create short video clips from images and prompts, with companies like Quora and Mondelez International already leveraging its capabilities. Quora plans to integrate Veo into its Poe chatbot platform, while Mondelez, the parent company of Oreo, intends to produce marketing content in collaboration with its agency partners.
Veo, first revealed in April, can generate 1080p video clips lasting up to six seconds, capturing various cinematic and visual styles, from landscapes to time-lapses. It also supports editing and refinement of previously generated content. Since its initial announcement, Google has enhanced Veo to meet enterprise requirements, improving its performance on Vertex AI. Customers can now create videos in 720p, with options for 16:9 landscape or 9:16 portrait aspect ratios.
According to Google, Veo excels in handling visual effects and certain aspects of physics, such as fluid dynamics, making it competitive with other leading video-generation models, including those from Adobe, Runway, and OpenAI. However, Veo is not without flaws, occasionally struggling with consistent object placement and realistic motion physics.
The model was trained on extensive video datasets, though Google remains tight-lipped about the specific sources of its training data. While the company suggests some YouTube content might have been utilized, it emphasizes adherence to agreements with creators and reliance on publicly available sources. Google’s use of public data for training aligns with its interpretation of fair use, though the practice has drawn criticism from some content creators concerned about copyright implications.
To address risks like regurgitation, where AI models replicate training data, Google has implemented prompt-level filters and a watermarking system called SynthID. Although not foolproof, SynthID embeds invisible markers into generated frames to help identify AI-created content.
Google has cautiously expanded Veo’s integration across its ecosystem, recently testing the model in Google Labs and incorporating it into YouTube Shorts for background and clip generation. Despite these developments, the broader adoption of Veo and its impact on creative industries remain to be seen, as Google continues refining the model for enterprise use. Meanwhile, its flagship image generator, Imagen 3, is now broadly available on Vertex AI, with additional features pending through a separate waitlist.
