[ad_1]
Google’s Video Creator is coming to a few customers — Google Cloud customers, to be exact.
Google announced Tuesday that Veo, its AI model that can create short videos from photos and prompts, will be available in private preview to customers using Vertex AI, Google Cloud’s AI development platform.
Google says the launch will enable one client, Quora, to bring Veo to its Poe chatbot platform, and another client, Oreo owner Mondelez International, to create marketing content with its agency partners.
“We built Poe to democratize access to the world’s best generative AI models,” said Spencer Chan, Poe’s product lead, in a statement. “Through partnerships with leaders like Google, we are expanding creative possibilities across all AI modalities.”
Leading generator
Unveiled in April, Veo can create 1080p clips of animals, objects and people up to six seconds long at 24 or 30 frames per second. Google says Veo is capable of capturing different visual and cinematic styles, including landscape and time-lapse shots, and making adjustments to already created shots.
Why the long wait for the API? “Enterprise readiness,” says Warren Barclay, senior director of product management at Google Cloud.
“Since announcing Veo, our teams have enhanced, strengthened and improved the model for Vertex AI’s enterprise customers,” he said. “Starting today, you can create Full HD videos at 720p resolution, in 16:9 horizontal or 16:9 vertical aspect ratios. Similar to how we have improved the capabilities of other models like the Gemini on Vertex AI, we will continue to do so from Yes Veo.
Veo understands visual effects well through prompts, Google says (think captions like “massive explosion”), and has a somewhat understanding of physics, including fluid dynamics. The model also supports masked editing to make changes to specific areas of the video, and is technically capable of stitching together footage into longer projects.
In these ways, Veo competes with today’s leading video creation models — not just OpenAI’s Sora, but models from Adobe, Runway, Luma, Meta, and others.
This 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 physics wrong. For example, cars will inexplicably and impossible to reverse on a dime.
Training and risks
Veo has been trained on a lot of shots. This is generally how generative AI models work: given example after example of some form of data, the models pick up patterns in the data that enable them to generate new data — videos, in Vue’s case.
Google, like many of its AI competitors, won’t specify exactly where the data to train its generative models comes from. When asked about Veo specifically, Barkley said only that the model “may” be trained on “some” YouTube content “per (Google’s) agreement with YouTube creators.” (Alphabet, Google’s parent company, owns YouTube.)
“Veo has been trained on a variety of high-quality video description datasets that are highly curated for safety and security,” he added. “Google’s core models are trained primarily on publicly available sources.”
New York Times reports in April revealed that Google Expanded terms of service last year in part to allow the company to leverage more data to train its AI models. Under the old terms of service, it was not clear whether Google could use YouTube data to create products outside the video platform. But this is not the case under the new conditions, which ease restrictions significantly.
While Google hosts tools to allow webmasters to prevent the company’s bots from scraping training data from their websites, it does not offer a mechanism to allow creators to remove their work from existing training sets. Google confirms that training models using publicly available data are… Fair useWhich means that the company believes that it is not obligated to seek permission from or compensate data owners. (However, Google says it does not use customer data to train its models.)
Thanks to the way today’s generative models behave when trained, they carry certain risks, such as regurgitation, which refers to when the model creates an exact copy of the training data. Tools like Runway’s have been found to spit Stills They are very similar to those found in copyrighted videos, setting up a potential legal minefield for users of the tools.
Google’s solution is Veo’s instant level filters, including violent and explicit content. If those fail, the company says Compensation policy Provides a defense to eligible Veo users against claims of copyright infringement.
“We plan to offset Veo outputs on Vertex AI when it becomes generally available,” Barclay said.
View everywhere
Over the past few months, Google has slowly built Veo into more of its apps and services as it works to improve the model.
In May, Google brought Veo to Google Labs, its early access program, for a select group of testers. In September, Google announced Veo’s integration of YouTube Shorts, YouTube’s short-form video format, to allow creators to create backgrounds and six-second videos.
What about the deepfakes dangers of it all, you might be wondering? Google says it uses its own watermarking technology, SynthID, to embed invisible tags into the frames Veo creates. SynthID is certainly not guaranteed against modifications, and Google has not made the Content ID portion available to third parties.
These could be controversial points if Veo doesn’t gain meaningful traction. On the partnerships front, Google has ceded ground to its AI creative rivals, who have moved quickly to lure producers, studios and creative agencies with their tools. Runway recently signed a deal with Lionsgate to train a custom model on the studio’s film catalog, and OpenAI has teamed up with Trademarks and Independent directors To showcase Sora’s potential.
Google at one point said it was exploring Veo apps in collaboration with artists including Donald Glover (AKA Childish Gambino). The company did not provide any update on these awareness efforts today.
Google’s pitch for Veo — a way to cut costs and quickly iterate on video content — runs the risk of alienating creators. A 2024 He studies Commissioned by the Animation Guild, a union representing Hollywood animators and cartoonists, it is estimated that more than 100,000 film, television, and animation jobs in the United States will be disrupted by artificial intelligence by 2026.
This may explain Google’s cautious “slow and steady” approach. When asked, Barclay did not provide an ETA for Veo’s general availability in Vertex, nor did he say when Veo might come to additional Google platforms and services.
“We typically release products for preview first, as this allows us to get real-world feedback from a select group of our enterprise customers before they become generally available for wider use,” he said. “This helps improve functionality and ensure the product meets our customers’ needs.”
In a related announcement today, Google said that its flagship image generator, Imagen 3, is now available to all Vertex AI customers with no waiting list. It’s gained new customization and photo editing features — but those features are locked behind a separate queue for now.
[ad_2]