OpenAI, Google and Runway race to generate AI video
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Sarah Grillo/Axios
With the wide release of Sora, OpenAI's video tool, most of the big tech giants — and some startups — are now racing to create models capable of generating realistic, high-quality videos from text prompts.
Why it matters: GenAI video tools could save time and money for filmmakers, but they could also unleash novel copyright issues and a flood of deepfakes.
- The fiercer the competition in the AI space, the more likely tech companies are to prioritize release dates over safety.
- Generating AI video requires magnitudes more energy than generating text, which is already straining the power grid.
The big picture: Google, OpenAI and AI startup Runway all offer products that let creators generate short videos with minimal effort.
Google's Veo 2
Google announced Veo 2, the latest version of its genAI video tool, a week after Sora's wider release in December. The company says Veo 2 is now available to early access users in the U.S. who are over 18.
- But good luck testing it yourself.
- "No one seems to have any idea how to actually try it out," tech writer and investor MG Siegler wrote in his blog. "I worked [at Google] for over a decade and I can't begin to tell you how you access it."
- Google hasn't released pricing details.
Between the lines: Google says Veo lets you create clips up to two minutes long, which means it could generate very short films.
- An entire section of Veo 2's product page is devoted to the tool's limitations, explaining that it struggles with consistency and complex motion.
- Ex-Googler Bilawal Sidhu had early access and posted a video of a dog wearing sunglasses and headphones and drinking a cocktail. Only the color of the umbrella in the cocktail glass changed in different shots.
- He also posted Veo 2 videos with uncanny human-like expressions using the prompt "An Indian uncle's reaction after you tell him you switched majors from mechanical engineering to sociology at a Christmas party."
While there's no agreed-upon benchmarking system for AI video generators, Google's Veo 2 is widely seen as the most promising right now.
- GenAI video creator Blaine Brown praised Veo by showing off videos of how different tools handled the prompt "a pair of hands skillfully slicing a perfectly cooked steak on a wooden cutting board. Faint steam rising from it."
- The test, Brown wrote on X, challenges AI models with physics and movement, an interpretation of what it means when a steak is cooked perfectly and depicting hands — a notorious struggle for AI.
Yes, but: There's a waiting list for Veo 2, while other models are available right now.
OpenAI's Sora
OpenAI released the beta version of Sora to a select group of testers in February 2024, then released the product to all ChatGPT Plus and Pro subscribers in December.
- OpenAI limits its $20/month Plus subscribers to 50 videos at 480p resolution or fewer videos at 720p per month.
- $200/month Pro subscribers are promised "10x more usage, higher resolutions, and longer durations."
- OpenAI CEO Sam Altman posted on X last month that all ChatGPT Plus users had unlimited access to Sora "over the holidays."
Between the lines: Sora is bundled with ChatGPT Plus and Pro, so the $20 or $200 subscription comes with all the text, coding and image-generating capabilities of ChatGPT Plus and Pro.
- Aimed at users who want to create 20-second videos, Sora is ideal for social media and marketing.
- Creating a video with Sora is as simple as entering a description in the prompt box and waiting for it to generate.
- The videos are impressive, but they often struggle with basic physics.
Fun fact: YouTuber Marques Brownlee's early review of Sora pointed out new and potentially problematic copyright issues with the tool.
- Brownlee asked Sora to create a video of a "tech reviewer sitting at a desk." The generated video featured a fake plant on the desk that is nearly identical to the plant Brownlee keeps on his desk for many of his tech reviews, suggesting the tool has been trained on his videos.
Runway's Gen-3 Alpha
AI startup Runway's most current model — Gen-3 Alpha — has been available on all of its paid plans since September.
- Runway's tools were critical to making the Oscar-winning film "Everything Everywhere All at Once," Axios reported.
- Subscriptions range from $144 to $1,500 per year depending on how many seconds of video you want to generate.
Between the lines: Runway was first out of the gate with the public release of its tool in 2023, but this doesn't always mean success. Think Betamax, Netscape, MySpace.
- Runway's longer tenure has given it more time to work with creators to prevent copyright infringement.
- The company inked a deal with Getty Images in December 2023 to offer a more "commercially safe" version of its tool, and partnered with Lionsgate Studios in September 2024 to train Runway's AI model on Lionsgate's movie catalogue.
- In December 2024, Runway launched a talent network for independent AI creators to show off their work to studios and production companies.
What we're watching: The race is still anyone's to win, including American giants like Adobe and Meta, Chinese tech companies like Tencent (Hunyuan Video) and Kuaishou (Kling AI), and a collection of other scrappy startups.
