Google has unveiled its latest video generation model, named Veo, which the company describes as its “most capable” to date. The model can generate high-quality, 1080p resolution videos exceeding 60 seconds in length and offers a variety of styles.
While Veo is not yet available to the general public, a limited number of creators will have access to it through an experimental tool called VideoFX. Additionally, Google has opened a waitlist for other interested users who wish to try the new technology.
To showcase Veo’s capabilities, Google released several video clips purportedly generated by the model. These included hyper-realistic aerial nature shots and whimsical scenes, such as a crocheted elephant walking and alpacas dancing.
Notably, many of the clips did not feature clear depictions of humans or their faces and bodies, and some exhibited an over-edited appearance indicative of their AI origins.
Google acknowledged that maintaining visual consistency in videos featuring humans can be challenging. The company plans to gather feedback from filmmakers and creators to improve Veo’s performance in this area.
“Veo accurately captures the nuance and tone of a prompt, providing an unprecedented level of creative control,” Google stated.
“It understands prompts for various cinematic effects, such as time lapses or aerial shots of a landscape.”
The introduction of Veo follows the release of OpenAI’s Sora text-to-video generator earlier this year.
Sora, which a limited group initially tested, produced sample clips, including a female model walking down a neon-lit city street. However, these clips also displayed issues characteristic of AI-generated content, such as warped movements and body parts.
Google emphasized that Veo was designed with responsibility in mind. Videos generated by Veo are watermarked using SynthID, a tool developed by Google for watermarking and identifying AI-generated content. The videos also pass through safety filters and memorization checks to mitigate privacy, copyright, and bias risks.