In June, Canva and Google worked together to bring VEO3's AI-generated audio and improved video to Canva Pro Masses. I got a demo of what Gen-AI tools can do.
Canva's push to the enterprise market is paying off in a huge way. Enterprise sales doubled year-on-year, COO Cliff Obrecht announced at Sydney's recent Google Cloud Summit. Currently, annual recurring revenue is USD 3.3 billion, up 43% from USD 2.2 billion in the second quarter of 2024.
The tech giant has long been on a rapid growth trajectory against IPOs. But like many successful scale-ups, AI is beyond the rise.
The video is a key component of Australia's Juggernaut's AI strategy, says Danny Woo, head of AI at Camba.
“Video is one of the fastest growing segments in Camba and is something we are deeply invested in,” Wu said. Forbes Australia.
Of course, to make a video sing, you need to use audio a lot. And up until now, no video generated by AI has been delivered on that front.
“The VEO3 is the first model to be synchronized and have excellent audio for video clips,” says Wu. “We've been integrating AI video creation for two years, and the VEO3 is unique in that it comes with quality output and sound.”
Place VEO3 in CANVA test
Wu provided us with a demonstration of Canva's new AI video generation tool via Google Meet. He asks what scenes I want to bring to life. Thinking about how he likes to spend his downtime, he asks if he can recall the horses running through the Victorian High Country.
“One of the things Canva allows is to customize the style. Get something very cinematic,” says Wu as he enters a prompt on his keyboard. “And from a lighting standpoint, Golden Hour. Run through the country alongside the Victorian Australian countryside. Add beautiful scenery to the background.”
“It's not really video without audio,” Wu says. I proudly look forward to the results. “And this is a very advanced and powerful model that currently takes just a minute or two for generations.”
While Canva is making 8-second videos, AI Ingeneration is deeply pondering, the conversation with WU turns into the engineering brain behind the advancements in audio and video.
“Like the transformers that created LLMS, much of the AI revolution journey came from Google researchers,” says Wu. “Google and Deepmind have an incredible history in their journey of AI and machine learning over the years. We at Canva are one of the global platforms offering this advanced VEO3 model.”
I ask him how much input Canva has in its development.
“We've been testing it on Google during the pre-release process and share feedback on which elements of the model worked well and which models need improvement,” says Wu.
One thing that continues to be refined is the time it takes to “delivery” user prompts. I think the results are already pretty fast.
Less than two minutes after giving the bucolic horse prompt, Wu returns to the screen and plays with the AI-generated video. As promised, it's like a cinema, the countryside scenery is beautiful, and the thoroughbred hooves slam the epic sound into the dirt. He said that audio was not specifically requested, but LLM provided what appears to be the best fit for the scene.
This clip needs kangaroos in the background, Wu tells me, and why not make three horses instead of two. I couldn't agree any more.
“The AI generation across all features is really good at it. It's repetition. So if this isn't what you're looking for, you can say, 'Make a similar clip, but black and white like a vintage movie,'” he says.
I agree that this particular video is most experienced in colour, but I agree that kangaroos and additional horses are mandatory. The prompt will take you within a minute and you will be ready for the second version. It's equally impressive.
Considering the potential applications of what I have shown, I ask the WU who holds copyright for what we have created.
“For commercial purposes, you can use everything created with Canva AI, including videos,” Wu helps. “As a customer and a camber, between you, you have all the rights to it. Therefore, you can use it commercially.”
Essentially, Google builds the technology, Canva delivers it to end users in easy-to-use packages, and Pro customers retain the rights to it as part of the benefits of their subscription.
Techniques Behind Video: Google's VEO3
Jason Gelman is the director of product management at Google's New York office. He has worked with WU repeatedly on providing VEO3.
“It creates fluid, natural videos that synchronize video with dialogue and video in a single pass, making it easy to capture creative nuances and detailed scene interactions at the prompt,” Gelman says of the third version of VEO.
Canva uses VEO3 on Vertexai, the AI platform of Google Cloud, and this product is now available to non-cumba users via the Google platform, finding traction worldwide.
The company says that since May, when VEO3 first became available, 70 million seconds of videos and nearly 50 million seconds of videos have been created.
Prompter protection
Because it is so widely used, it was important to incorporate safeguards like digital watermarks and safety filters into the VEO3, says Gelman. Content generated in sync with the safety filter is covered by Google's Gen AI compensation. This is a $3.5 trillion tech promise that if GEN AI users are “challenging copyright grounds,” Google will be liable for potential legal risks.
Canva also has built-in protection. Their “shield” promise compensates eligible enterprise customers who use Magic Studio to generate AI content.
“In true Canva fashion, we built this with an intuitive interface and simple editing tools in place, all underpinned by Canva Shield,” Canva CTO Cameron Adams said when the VEO3 Canva AI collaboration was announced. “[It’s] A genuine step change of creativity. ”
Look back at the week with carefully selected articles from Australia and around the world. Sign up for the Forbes Australia newsletter Or become a member here.
