OpenAI, the company behind ChatGPT, has unveiled a brand-new AI model that is astounding. The model, named Sora, can record a full minute of video using only text commands. According to the OpenAI Sora blog, we’re teaching AI to comprehend and mimic the physical world in motion with the intention of developing models that assist people in solving issues that call for real-world interaction.
Sam Altman, the CEO of OpenAI, also showcased the tool through posts on his X account. We want to show you what Sora is capable of. Please respond with the titles of the videos you would like to see, and we will get to work creating them! As stated in Altman’s post. He received a lot of prompts from users on the platform, and the outcomes he shared seem extremely accurate.
It’s a little difficult to believe how real this Sora-generated video looks, as shared by another OpenAI member.
According to OpenAI, Sora is capable of creating complex scenes with numerous characters, accurate movements, and elaborate backgrounds. Along with understanding user prompts, the model also deciphers how these elements appear in real-world scenarios. The model’s profound linguistic comprehension allows it to reliably decipher prompts and produce engrossing characters that vividly convey emotions. According to OpenAI, Sora can also produce multiple shots that faithfully maintain characters and visual style within a single generated video.
Although the OpenAI Sora model has generated a lot of excitement online this morning, well-known YouTuber Marques Brownlee, also known as MKBHD, has correctly brought up some potential issues with the model. “All of these videos are artificial intelligence (AI)-generated, and nothing will worry you more than a little bit about this,” he stated in his post.
The OpenAI Sora model’s current capabilities certainly seem impressive. But, it’s important to be cautious of models that can easily produce a one-minute video from simple text prompts, as they may be abused. You’ll be able to distinguish Sora’s significant advancement from the viral AI-generated video of Will Smith eating spaghetti that went viral in early 2023.
Nonetheless, OpenAI is not avoiding the topic of safety. According to the company, they will be putting in place a number of vital safety precautions before releasing Sora in OpenAI’s products. This entails working with red teamers, specialists in fields like bias, hate speech, and disinformation, who will thoroughly test the model to find any potential weaknesses. Furthermore, OpenAI will be creating tools to identify deceptive content, like a detection classifier that can recognize videos produced by Sora.
Additionally, OpenAI will make use of DALL·E 3 safety protocols that have already been established for its products, which also apply to Sora. For example, the text classifier in an OpenAI product will filter and reject input prompts that don’t follow usage guidelines, like those that call for extreme violence, explicit sexual content, hateful imagery, or the likeness of famous people. According to the company, it has also built strong image classifiers that examine each frame of produced videos to make sure our usage guidelines are followed before granting users access.
OpenAI also says that it is actively engaging policymakers, educators, and artists globally to address concerns and explore positive applications of this new technology. “We intend to involve policymakers, educators, and artists globally in order to comprehend their apprehensions and ascertain constructive applications for this novel technology. We cannot foresee every constructive or harmful way that people will use our technology, even after conducting a great deal of research and testing. Because of this, we think that developing and releasing ever-more-safe AI systems over time requires learning from real-world applications.” says OpenAI.
Sora can only be used by red team members to scan key areas for possible problems or hazards at this time. Additionally, OpenAI is making its services available to designers, filmmakers, and visual artists in order to get their feedback on how to make the model better.