OpenAI has recently released a new text-to-video generator called Sora, which was created by ChatGPT. Here is the information we currently have about this new tool.
The company behind ChatGPT is expanding its reach to include AI-generated videos.
Introducing Sora – a new text-to-video creator developed by OpenAI. The platform, revealed by the San Francisco-based company on Thursday, utilizes generative AI to swiftly generate brief videos based on written instructions.
Other individuals have also showcased similar technology before Sora. However, experts in the field highlight the exceptional level of video quality produced by the tool thus far. They also acknowledge that its release signifies a major advancement for both OpenAI and the advancement of text-to-video generation overall.
However, like many developments in the rapidly expanding field of AI, this technology also triggers concerns about ethical and societal consequences. Here is the essential information you should be aware of.
Sora is a text-to-video generator — creating videos up to 60 seconds long based on written prompts using generative AI. The model can also generate video from an existing still image.
Generative AI is a branch of AI that can create something new. Examples include chatbots, like OpenAI’s ChatGPT, and image-generators such as DALL-E and Midjourney. Getting an AI system to generate videos is newer and more challenging but relies on some of the same technology.
Unfortunately, Sora is not currently accessible to the public as OpenAI is in discussions with government officials and creatives before officially launching the program. There is still a significant amount of information that has not been revealed, but the company has provided a few demonstrations of videos created by Sora since the announcement on Thursday.
Sam Altman, the CEO of OpenAI, used X (formerly known as Twitter) to request ideas for prompts from social media users. He then shared videos that realistically depicted scenarios such as two golden retrievers podcasting on top of a mountain and a bicycle race on an ocean with various animals riding the bikes and captured from a drone’s perspective.
Although Sora-generated videos are capable of portraying intricate and highly detailed scenarios, OpenAI acknowledges that they still have some flaws, particularly in regards to spatial and causal relationships. For instance, as stated on their website, “a person may appear to take a bite out of a cookie, but the cookie may not actually have a bite mark afterwards.”
Sora by OpenAI is not the initial technology of its kind. Other companies such as Google, Meta, and Runway ML have also showcased comparable capabilities.
However, experts in the industry highlight the perceived high quality and considerable duration of Sora videos that have been shared thus far. Fred Havemeyer, the leader of Macquarie’s research on AI and software in the United States, stated that the release of Sora signifies a significant advancement for the sector.
According to Havemeyer, the new feature allows for longer videos, possibly up to 60 seconds. These videos also appear more realistic and adhere to the laws of physics and the real world. As a result, there are fewer “uncanny valley” or odd-looking videos on the feeds.
There have been significant advancements in AI-generated videos in the past year, including the release of Stable Video Diffusion in November. However, Forrester senior analyst Rowan Curran notes that these videos still require manual editing for consistency in characters and scenes.
According to Curran, the frequency and duration of Sora’s videos offer new possibilities for artists to integrate AI-generated elements into traditional content. They can also now create complete narrative videos based on just a few prompts. Curran shared this information with The Associated Press via email on Friday.
Despite Sora’s remarkable abilities demonstrated since its launch on Thursday, concerns about the ethical and societal impact of using AI-generated videos still persist.
Havemeyer highlights the significant dangers in the upcoming election cycle of 2024, for instance. He also mentions that having a potential tool to create videos that appear and sound realistic could lead to various problems in the realm of politics and beyond. He specifically mentions concerns about fraud, propaganda, and misinformation.
Havemeyer stated that the repercussions of generative AI will be a significant subject for discussion in 2024. This is a significant problem that will affect all businesses and individuals in the coming year.
Technology corporations are maintaining control over the regulation of artificial intelligence and its potential dangers, while governments globally strive to keep pace. In December, the European Union came to an agreement on the first all-encompassing regulations for AI, but it will not be implemented until two years after receiving final approval.
OpenAI announced on Thursday that they are implementing significant safety measures before releasing Sora to the public.
“We are collaborating with a team of experts in fields such as misinformation, harmful content, and prejudice – known as the ‘red team’ – to conduct adversarial testing on our model,” stated the company. “Additionally, we are developing tools to identify deceptive content, including a classifier that can determine if a video was created by Sora.”
During her speech at the Munich Security Conference, Anna Makanju, OpenAI’s Vice President of Global Affairs, emphasized the importance of addressing AI-generated election deepfakes. Along with 19 other tech companies, OpenAI has committed to collaborating voluntarily to combat this issue. Makanju also mentioned that the company’s release of Sora will be done with caution.
OpenAI has not provided much detail on the development of Sora. Their technical report did not mention the specific imagery and video data used to train Sora, and the company has not yet responded to a request for additional information as of Friday.
The release of Sora comes at a time when OpenAI and Microsoft are facing legal action from authors and The New York Times for using copyrighted written works to train ChatGPT. OpenAI compensates the AP for the use of their text news archive, but the amount paid remains undisclosed.
___
O’Brien reported from Providence, RI.
Source: wral.com