(Credit: mikkelwilliam / E+ via Getty Images)
The generative AI craze began with ChatGPT's text-based responses, but has evolved to include image generation and now text-to-video.
Four such tools launched this week, from Amazon, YouTube, Alibaba, and movie studio Lionsgate. They create videos from written descriptions (think: "Create a video of a dog catching a ball"), but the tech still has a long way to go.
YouTube's Veo tool can only create a green-screen-style background for its vertical clips, known as Shorts. Complete video clips (up to six seconds in length) will follow in 2025.
Amazon's tool helps sellers create simple advertisements—really simple. An example clip brings to life a static image of a coffee mug, and makes a video of it steaming against an artificial background. Details about Alibaba's tool are slim, but the announcement happened alongside the release of 100 open-source models, CNBC reports.
These tools pale in comparison to what OpenAI is promising with its Sora video generator, which can generate up to 60-second silent clips. (Google made similar promises for Veo at I/O in May.) When Sora debuted in February, it shocked the internet with hyper-realistic videos. Toys R Us even used it to whip up a commercial with a full storyline and flawless animation.
Sora is not yet available to the public. OpenAI CTO Mira Murati says it's coming at this end of this year, with sound generation to follow "eventually."
In the meantime, OpenAI has focused on releasing two AI models: GPT-4o and GPT-o1. The latter, known as "Strawberry," is primarily geared toward scientists and engineers who need to solve complex reasoning problems.
With video, Amazon and YouTube have beaten OpenAI to the punch when it comes to practical tools for the public. Even if Sora is technically "better" at creating videos—and that remains to be seen—Amazon and YouTube could get greater usage out of theirs since they are embedded in sites that millions of people use every day. Plus, we don't know how much Sora will cost, given the extra high computational demands, and what user limits will accompany it.
Seeing Is Believing?
Are we sure we want more AI videos? They heighten existing concerns about AI deepfakes and could destroy our ability to believe what we see online.
Historically, video has been the ultimate source of truth. Emails, texts, and photos can be doctored in Photoshop, but a video? That's hard to refute. It's the ultimate gift to those who seek to disparage or misrepresent someone by creating a "video" of them. Though there has been at least one positive use for AI videos, by journalists in dictatorial Venezuela, the potential for bad actors to abuse these systems could outweigh the positives.
Lionsgate's new partnership with AI startup Runway could save the Hunger Games creator "millions and millions of dollars," Vice Chairman Michael Burns tells The Wall Street Journal. Amazon similarly touts its AI video generator as a cost-saving measure for small businesses that cannot hire costly production crews.
Burns says Runway's model will augment creative work, not replace it. Others argue the tech threatens jobs for storyboard and VFX artists. The company also offers static image generation and has been sued by artists for copyright infringement, Artnet reports.
Perhaps OpenAI is moving slow with Sora's public release since it gives the company time to test it and implement safeguards. In February, the company promised to take "several important safety steps ahead of making Sora available in OpenAI’s products, [like] working with red teamers—domain experts in areas like misinformation, hateful content, and bias—who are adversarially testing the model."
In the US, we have only a patchwork of laws to address this type of technology. Profit is still the ultimate incentive. Some tools, like Google Gemini and Midjourney, have put restrictions in place around public figures like presidential candidates. However, Elon Musk's Grok AI allows image generation of famous people, no matter how bizarre the request. No regulation, with huge profit incentives—what could go wrong?