Sora, OpenAI's video generator, was released today to at least some users.
YouTuber Marques Brownlee revealed the news in a video posted to his channel on Monday morning. Brownlee got early access to Sora and gave his first impressions in a 15-minute review.
According to Brownlee, Sora lives at Sora.com, whose home page features a scroll of recently generated Sora videos curated by OpenAI. (As of this article's publication, it had not yet been published on TechCrunch.) Notably, the tool is not built into ChatGPT, OpenAI's AI-powered chatbot platform. Sora seems to be its own separate product for now.
Videos on the Sora homepage can be bookmarked to the Saved tab for later viewing, organized into folders, and clicked to see which text prompts were used to create them. According to Brownlee, Sora can generate videos from uploaded images and prompts, and you can also edit existing videos created with Sora.
Using the “Remix” feature, users can write the changes they want to see in the video, and Sora will try to incorporate them into the newly generated clip. Remix has a “strength” setting that allows users to specify how drastically Sora changes the target video. Higher values produce videos with more artistic freedom.
Brownlee said Sora can produce up to 1080p footage, but the higher the resolution, the longer it takes to produce the video. 1080p footage takes eight times longer than the fastest option, 480p, while 720p takes four times as long.
Visit Sora's homepage. Image credit: Marques Brownlee
Brownlee said that in testing, it took “several minutes” to generate an average 1080p video. “It's also at a time when almost no one else is using it at the moment,” he said. “I wonder how long it will take until everyone can use this.”
Brownlee said that in addition to generating one-off clips, Sora has a “storyboard” feature that allows users to combine prompts to create sequences of scenes and videos. This is probably meant to increase consistency, which is a notorious weakness of AI video generators.
But how does Sora perform? According to Brownlee, this tool suffers from the same flaws as other generation tools, namely issues related to object persistence. In Sora videos, objects pass in front of or behind each other in incomprehensible ways, disappearing and reappearing for no reason.
Brownlee said Sola's feet are also a big problem. Whenever a person or animal with legs has to walk for a long time in a clip, Sora confuses front and hind legs. The legs “switch” back and forth in a way that is anatomically impossible, Brownlee said.
Sora's terms of use. Image credit: OpenAI
Brownlee said Sora has a number of safeguards built in to prevent creators from showing people under 18, containing violence or “explicit themes,” and potentially infringing third-party copyrights. It is said that it is prohibited to create videos that contain. According to Brownlee, Sora does not generate videos from images containing celebrities, recognizable characters or logos and watermarks each video, but visual watermarks can be easily cropped. .
So what is Sora useful for? Brownlee found this useful for certain styles of title slides, animations, summaries, stop-motion footage, and more. However, he stopped short of endorsing realism.
“It’s impressive that it’s an AI-generated video, but you can tell right away that it’s an AI-generated video,” he said of most of Sora’s clips. “Things are getting really volatile.”