OpenAI’s much-hyped video generator, Sora, finally debuts, but overwhelming demand and limited availability highlight the scaling challenges generative AI firms face as they race to deploy.
After months of anticipation and a rocky preview phase, OpenAI on Monday finally launched Sora, its cutting-edge AI video generator—but overwhelming demand quickly sapped momentum. Teething issues aside, the launch solidifies the company’s position as a leader in generative AI, as firms race to advance and deploy the fast-evolving technology, while society navigates thorny concerns around safety and protecting creative jobs.
The Microsoft-backed startup, recently valued at nearly $160 billion, said it was moving the video generation model out of research preview as part of its 12-day “ship-mas” series of product launches, though the rollout was stymied by high demand, prompting the company to slow video generations and disable signups.
What Can OpenAI’s Sora Do?
The product, Sora Turbo, is a newer, faster version of the Sora model OpenAI released to a group of creatives for testing in February.
The tool can be used to generate videos based on text, image or video prompts and can also be used to “extend, remix, and blend” users’ own assets. OpenAI said users can use Sora to generate clips up to 1080p resolution, up to 20 seconds long and in a variety of aspect ratios, including widescreen, vertical and square.
Sora Turbo will be available at no extra cost to ChatGPT Plus and Pro users, though there will be limits on how many videos users can create and at what quality depending on the user’s plan. ChatGPT Plus users can generate up to 50 videos at 480p resolution a month, or fewer at 720p, OpenAI said, with Pro users getting ten times the usage, higher resolutions and longer durations. In a post on X, OpenAI CEO and cofounder Sam Altman said Pro users can get 500 fast generations a month, or fewer at high resolution, and an unlimited number of slower generations.
“We’re working on tailored pricing for different types of users, which we plan to make available early next year,” OpenAI said in a blog post.
Throttled by High Demand
Sora, which comes to market a week after Google released video generator Veo to cloud customers and competes with products from Meta and Runway, will be available as a standalone tool at Sora.com. It will debut in the US and “everywhere ChatGPT is available,” barring the UK, EU countries and Switzerland. OpenAI did not provide a reason for excluding these countries, nor a timeline for Sora’s release there, though tech companies have often voiced concerns over the strict regulatory environment in these countries.
“We are working to expand access further in the coming months,” OpenAI said.
The Sora rollout was also marred by intense demand for the product, company executives said, leading to slower generation speeds and the decision to pause signups for the service. “We significantly underestimated demand for Sora,” Altman wrote on X. “It is going to take a while to get everyone access.”
“Trying to figure out how to do it as fast as possible!”
Combating Deepfakes and Misinformation
Generative AI tools capable of creating realistic video, audio and images have sparked particular concern among regulators and activists due to their ability to be used to facilitate misinformation and abuse. The potential use of tools like Sora to craft deepfakes, which can be convincing but fake content of real people, are of particular concern and leaders like OpenAI, Google and Microsoft have come under intense scrutiny to ensure there are adequate safeguards in place to minimise the risk of abuse.
OpenAI acknowledged it was deploying a version of Sora that “has many limitations,” such as generating unrealistic physics or struggling with complex actions over long durations, but said it was doing so with safety in mind. The decision to release a product with flaws was to “give society time to explore its possibilities and co-develop norms and safeguards that ensure it’s used responsibly as the field advances,” OpenAI said.
In addition, OpenAI said it was taking steps on its end to ensure Sora is used responsibly, including blocking sexual deepfakes and child sexual abuse materials and adding metadata and watermarks to identify content as AI-generated video. Uploads of people will also be limited at launch, OpenAI said, though restrictions will be relaxed “as we refine our deepfake mitigations.”
Creative Outcry
Video and image generators like Sora, as well as rival AI products from the likes of Google, Microsoft, Runway and Stability, have also sparked outcry from creative professionals, many of whom claim AI models are trained on stolen data and put their livelihoods at risk.
Sora’s release comes months an early version of the model was handed to a select group of artists and other creatives to test earlier this year. The model was reportedly pulled from testing in November after a group released a demo version and attacked OpenAI for allegedly exploiting artists for “art washing” and unpaid labour.
OpenAI rejects these claims and says testers were neither required to use the product nor give feedback on it. The controversy underscores ongoing tensions as generative AI is rolled out across creative industries and commercial uptake will be hindered until unresolved legal and policy issues surrounding copyright and AI models are resolved.