Google's video generator is coming to a few more customers, specifically Google Cloud customers.
Google announced Tuesday that Veo, an AI model that can generate short video clips from images and prompts, will be available in private preview for customers using Vertex AI, Google Cloud's AI development platform.
Google says the launch will enable one of its customers, Quora, to integrate Veo into its Poe chatbot platform, and another customer, Oreo owner Mondelez International, will be able to integrate Veo into its Poe chatbot platform. It will be possible to create marketing content with agency partners.
“We created Poe to democratize access to the world's best generative AI models,” Spencer Chan, Poe's head of product, said in a statement. “Through partnerships with leaders like Google, we are expanding creative possibilities across all AI modalities.”
flagship generator
Veo, announced in April, can produce 1080p clips of animals, objects, and people for up to six seconds at 24 or 30 frames per second. Google says Veo can capture a variety of visual and cinematic styles, such as landscapes and time-lapse shots, and edit already generated footage.
Why are my APIs experiencing high latency? Warren Barclay, senior director of product management at Google Cloud, says it's “ready for the enterprise.”
Some of Veo's works. Image credit: Google
“Since Veo was announced, our team has enhanced, enhanced, and improved models for enterprise customers on Vertex AI,” he said. “As of today, you can create high-resolution videos in 720p, 16:9 landscape or 9:16 portrait aspect ratio. Just as we have improved the capabilities of other models such as Gemini on Vertex AI, Veo But we will continue this.”
According to Google, Veo understands VFX pretty well from the prompts (think captions like “giant explosion”) and also has some understanding of physics, including fluid mechanics. This model also supports mask editing to change specific areas of your video, and technically allows you to splice footage into longer projects.
In this way, Veo can compete with today's leading video generation models such as OpenAI's Sora as well as Adobe, Runway, Luma, and Meta.
If you provide a reference image with a prompt, Veo will generate a video according to the style of the image and the instructions in the prompt. Image credit: Google
That doesn't mean Veo is perfect. Reflecting the limitations of today's AI, objects in Veo's videos disappear and reappear without much explanation or consistency. And Veo often gets its physics wrong. For example, the car suddenly backs up in an inexplicable and impossible way.
training and risks
Veo was trained with a lot of footage. This is how generative AI models generally work. If you provide example after example of some form of data, the model can recognize patterns in the data and generate new data (video in Veo's case).
Like many AI rivals, Google doesn't reveal exactly where it sources the data to train its generative models. Asked specifically about Veo, Barkley said only that the model “could be trained” on “some” YouTube content. [Google’s] Agreement with YouTube Creators. (Google's parent company Alphabet owns YouTube.)
“Veo is trained on a variety of high-quality video explanation datasets, carefully selected for safety and security,” he added. “Google's underlying models are trained primarily on publicly available sources.”
An April report in the New York Times revealed that Google expanded its terms of service last year to allow it to use more data to train its AI models. Under the old terms of service, it wasn't clear whether Google could use YouTube data to build products beyond its video platform. The new conditions do not, and the reins are loosened considerably.
Another sample from Veo. Image credit: Google
Google hosts tools that allow webmasters to block its bots from scraping training data from websites, but it doesn't provide a mechanism for creators to remove their work from existing training sets. Google claims that training models using publicly available data is fair use. This means that the company believes it is under no obligation to ask permission or compensate data owners. (However, Google says it doesn't use customer data to train its models.)
Because of the way today's generative models operate during training, they come with certain risks, such as backflow when the model produces a mirror copy of the training data. Tools like Runway have been found to spit out still images that are substantially similar to copyrighted videos, creating a potential legal minefield for users of the tools.
Google's solution is a prompt-level filter for Veo, which includes violent and explicit content. Should these fail, the company says its indemnification policy will protect eligible Veo users from claims of copyright infringement.
“Once Vertex AI becomes generally available, we plan to compensate Veo outputs,” Barkley said.
Anywhere Veo
Over the past few months, Google has been slowly incorporating Veo into more apps and services as it refines its model.
In May, Google introduced Veo to Google Labs, an early access program for select testers. And in September, Google announced Veo integration for YouTube Shorts, YouTube's short video format, to allow creators to generate backgrounds and six-second video clips.
You may be wondering, what about the risks of all these deepfakes? Google said it uses its proprietary watermarking technology, SynthID, to embed invisible markers in the frames Veo generates. Admittedly, SynthID is not completely secure against editing, and Google does not make the Content ID portion available to third parties.
Yet another sample from Veo. Image credit: Google
These may be moot if Veo doesn't gain meaningful traction. On the partnership front, Google is ceding ground to its generative AI rivals, who are quickly wooing producers, studios, and creative agencies with their tools. Runway recently signed a deal with Lionsgate to train a custom model on the studio's movie catalog, and OpenAI is working with brands and independent directors to showcase the potential of Sora.
Google at one point said it was working with artists including Donald Glover (aka Childish Gambino) to explore applications for Veo. The company did not release any updates regarding these support efforts today.
Google's pitch for Veo – a way to cut costs and quickly iterate on video content – risks alienating creatives. A 2024 study commissioned by the Animation Guild, a union representing Hollywood animators and cartoonists, found that more than 100,000 U.S.-based film, television and animation jobs will be destroyed by AI by 2026. It is estimated that
That may explain Google's cautious “slow and steady” approach. When asked, Barkley did not provide a timeline for general availability of Veo on Vertex, nor did he say when Veo would be available on additional Google platforms or services.
Image credit: Google
“We typically release products in preview versions first, as this allows us to get real feedback from a select group of enterprise customers before making them widely available to the public,” he said. Said. “This improves functionality and ensures the product meets customer needs.”
In a related announcement today, Google announced that its flagship image generator, Imagen 3, is now available to all Vertex AI customers without a waiting list. New customization and image editing features have been added, but these are gated behind a separate waiting list for now.
TechCrunch has a newsletter focused on AI. Sign up here to get it delivered to your inbox every Wednesday.