The latest iterations of Google's video generation AI model Veo 3 continue to evolve with rapid clips.
As part of the latest upgrade, this model allows users to generate 8-second video clips with audio generated by AI from a single still image. According to Google Cloud documentation updated on Monday, the feature is now available as a “preview product.” Josh Woodward, head of Google Labs and Gemini App, wrote in a post on X last week that the company is working on the inter-image feature of Veo 3.
What uses VEO 3
For example, an influencer can upload one headshot of himself and encourage the model to generate a short clip of him walking down the runway wearing a product from a partner brand. VEO 3 automatically contains ambient noise, like crowd tweets and footsteps on the floor. Users can also request that AI-generated portraits speak a few lines, as in this example.
Also: This new AI video editor is an all-in-one production service for filmmakers – how to try it
Brands can also use new features by feeding the model an image of the product and requesting clips to display from different perspectives. Amazon has developed an AI tool for advertisers with similar capabilities, but Meta vows to go further, stating its plans to automate the entire ad production process.
VEO 3's new image-to-video feature helps creative professionals from various industries save time and resources otherwise spent organizing video shoots on-site. It can also provide more creative material for use across social media and other channels.
Google unveiled VEO 3 at its annual I/O Developer Conference in May. This model quickly attracted attention from AI researchers and creative experts because of its ability to seamlessly integrate AI-generated video and audio. It is also excellent at simulation of real-world physics, not hampered by many technical flaws that plagued the video tools that were previously generated by AI.
Also: How AI companies secretly collect training data from the web (and why it's important)
There is no indication that Google's investment in VEO 3 will slow anytime soon. Last week, Google Deepmind CEO Demis Hassabis appeared to suggest that the model could be used immediately to generate a virtual world of video games. The timing of that prediction is interesting given that Microsoft fired 9,000 people from the gaming division earlier this week.
How to try it
Originally only available from Gemini Ultra and Flow, Veo 3 was generally released as a public preview last month. All Google Cloud customers and partners are accessible through Vertex AI Media Studio. The model is now available in 159 countries.
Controversy and potential risks
VEO 3 has recharged the spread of misinformation online, raising concerns about the possibility of AI interacting with users on social media. There are also questions about procuring training data, but Hassabis says it can include YouTube videos.
Also: This open source bot blocker shields your site from nasty AI scrapers – this is how
AI companies have slashed out much of the text, images, audio and video content they use to train models from the open internet, and creators across the publishing, art and film industries have raised copyright issues with these generators. If you're looking for video tools generated for more airtight AI, consider checking out Marey from Moonvalley, who claims to be trained only with licensed data.
