OpenAI released its video generation tool Sora on Monday. However, the company has chosen not to release any significant features to most users until further testing is conducted.
The feature in question generates a video using an uploaded photo or footage of a real person as a reference. OpenAI says it will give “some” Sora users access to Sora users, but won't widely roll out the feature until it has had a chance to fine-tune its “approach to safety.” There is.
“The ability to generate videos using uploaded photos or videos of real people as 'seeds' is a potential exploit vector, and we are looking to develop a particularly progressive approach to learn from early usage patterns. “We are taking a similar approach,” OpenAI wrote in the article. Blog post. “Early feedback from artists shows that they value this as a powerful creative tool, but given the potential for abuse, we have made it available to all users from the outset. It does not mean that
OpenAI also does not allow you to share generated videos that contain clips or images of real people to Sora's homepage discovery feed.
Generative video is a powerful tool, but it's also a controversial one, with deepfakes and misinformation being major concerns. According to data from ID verification service Sumsub, deepfake fraud worldwide increased more than 10 times from 2022 to 2023.
Among other steps OpenAI is taking to prevent abuse, Sora has a filter that detects whether the videos it generates depict people under the age of 18. If so, OpenAI will apply “tighter thresholds” for moderation regarding sexual, violent, or self-harmful content, the company claims.
All Sora-generated videos contain metadata that indicates their origin, specifically metadata that complies with C2PA technical standards. Metadata can be removed or added. But OpenAI is touting it as a way for platforms that support C2PA to quickly detect whether a video is from Sora.
OpenAI also says it is conducting “instant rewrites” to prevent Sora from producing videos in the style of active creators to fend off claims of copyright infringement.
“We've added a prompt rewrite feature designed to trigger when a user attempts to generate a video in the style of a living artist,” the company wrote. “We decided to take a conservative approach with this version of Sora to learn more about how Sora is used in the creative community… There's a very long tradition of building, but some creators may have concerns.
A number of artists are suing AI companies, including OpenAI, for training their work without permission to create AI tools that regurgitate content in their own unique styles. Both companies argue that the fair use doctrine protects them from claims of copyright infringement, and that in reality their AI models cannot be reverted.
According to video blogger Marcus Brownlee, who got an early preview of the feature, Sora can create videos from text prompts and images, as well as edit existing videos through a remix tool. The storyboard interface allows users to create a series of videos. The Blend tool, on the other hand, takes two videos and creates a new video that retains elements of both.
Subscribers to OpenAI's ChatGPT Pro and Plus plans can access Sora, OpenAI's video generator, starting today, but only if they reside in certain countries.