Last month, Sam Altman’s artificial intelligence startup OpenAI showed off Sora, its groundbreaking text-to-video generator. Now the Microsoft-backed AI company has revealed that it will release a public version of the tool later “this year”.
In an interview with the Wall Street Journal, OpenAI’s chief technology officer Mira Murati said Sora will be available to the general public within a “few months”. However, that didn’t stop Sora-generated videos from making their way onto social media platforms like X and Instagram.
OpenAI to Release Hyperrealistic Text-to-Video Generator Tool ‘Sora’ Soon
Sora is capable of generating hyperrealistic scenes from just text prompts. Until now, the tool has only been made available to media professionals like visual artists, designers, and filmmakers.
Apart from releasing the tool to the public, OpenAI also has plans to “eventually” incorporate audio, which according to the company, has the potential to make scenes generated by Sora sound and feel even more realistic.
Murati did not give an exact date as to when the feature will be available on Sora, but she said OpenAI is currently working on adding the functionality. The company also wants to add an editing feature to the tool that will allow users to edit and correct the generated content.
The CTO told the Journal that AI models like Sora are not able to always create accurate images and therefore including that feature can be a significant improvement. Murati noted that OpenAi is working on how to use the technology as a tool for users to edit and create content.
Sora Expected to be Priced the Same as OpenAI’s Text-to-Image Model DALL-E
When pressed about what data was used to train Sora, Murati responded with vague answers by not being specific and trying to dodge the question. She stated the model leveraged data that was publicly available and licensed.
It also remains uncertain whether the video contents used in training were sourced from YouTube, Facebook, and Instagram. However, Murati confirmed that Sora only uses image content from Shutterstock, which is a partner of OpenAI.
Another important point she highlighted was that Sora is much more expensive to operate, compared to OpenAI’s previous AI models like DALL-E or ChatGPT. The company aims to make the tool available to the general public at “similar costs” to DALL-E.
Murati also noted that, unlike its text and image-generation model, Sora is still a work in progress and a research project that requires a piece of specialized computing equipment before being released to the public, making it seemingly more expensive.
Concerns Mounting Over AI’s Ability to Create Misinformation and Influence Elections
As the 2024 US Presidential election nears, there is growing concern among regulators that generative AI tools can be used to promote misinformation – the frequency of which has only increased over the past year.
Murati said that it’s highly unlikely that Sora will be able to produce images of public figures, which will be in line with DALL-E’s policies. It is being reported that Sora-generated videos will have a watermark to distinguish them from the real ones.
Last week, Google announced that it would be limiting its large language model (LLM) Gemini’s ability to answer election-related questions. Meanwhile, Facebook is setting up an expert team of analysts to combat misinformation leading up to the European Union elections in June.
Representatives from OpenAI met with officials from the Election Committee of India last month to discuss how to prevent the misuse of ChatGPT in upcoming elections.
According to the World Economic Forum, 2024 is poised to be the single-largest election year in history with nearly 2 billion people from 64 countries expected to vote. The US, UK, and India – the world’s largest democratic nation – are set to host general elections this year.
Related News: Apple Acquires AI Startup In Push To Deliver On The AI Front And Compete With Google And Microsoft
