OpenAI on Thursday announced a tool that can generate videos from text prompts.
The new model, called Sora after the Japanese word for “sky,” can create up to a minute of realistic footage that follows the user’s instructions for both subject matter and style. The model can also create videos based on still images or enhance existing footage with new material, according to a company blog post.
“We teach AI to understand and simulate the physical world in motion, with the goal of training models that help people solve problems that require real-world interaction.” says the blog post.
One video included among the company’s first few examples was based on the following prompt: Movie trailer featuring the adventures of a 30-year-old astronaut wearing his red woolen knitted bike in his helmet, blue sky, salt desert, cinematic style shot on 35mm film, vibrant colors .”
The company announced that it has opened up access to Sora to several researchers and video creators. According to the company’s blog post, experts have “red-teamed” the product and implemented OpenAI’s terms of service, which prohibit “extreme violence, sexual content, hateful images, likenesses of celebrities, or the IP of others.” We will test whether there is a possibility of evasion. The company only allows limited access to researchers, visual artists and filmmakers, but CEO Sam Altman took to Twitter after the announcement to answer questions from users about a video he said was created by Sola. posted. The video contains a watermark indicating that it was created by AI.
The company debuted its still image generator Dall-E in 2021 and its generated AI chatbot ChatGPT in November 2022, quickly gaining 100 million users. His other AI companies have also debuted video generation tools, but those models could only generate a few seconds of footage that had little to do with the prompt. Google and Meta said they are developing a video generation tool, although it is not publicly available. on wednesday, announced the experiment We’ve added deeper memory to ChatGPT to remember more of your users’ chats.
OpenAI told the New York Times how much footage was used to train Sora, except that the corpus includes videos that are publicly available and licensed from copyright holders. He also did not reveal the source of the training video. The company has been sued multiple times for alleged copyright infringement in training generative AI tools that digest vast amounts of material collected from the internet and mimic the images and text contained in those datasets. .
Source: www.theguardian.com
