OpenAI introduces Sora, a new text-to-video model
OpenAI has announced the development of Sora, an advanced text-to-video model designed to understand and simulate real-world interactions through AI-generated videos. Sora is capable of creating videos up to a minute long while adhering to specific user prompts, maintaining high visual quality throughout.
The models website shows many impressive examples that highlights Sora’s ability to generate detailed, dynamic scenes based on descriptive prompts. OpenAI point out that in this early phase of development, it’s not without it’s limitations, it has some issues with accurately simulating complex physical interactions and specific cause-and-effect scenarios.
According to OpenAI, to ensure safety and ethical use they are collaborating with “red teamers” and a selection of designers and artsts to identify and mitigate potential risks. This includes testing for misinformation, biased content, and hateful material. Additionally, OpenAI plans to incorporate detection classifiers and C2PA metadata to distinguish AI-generated videos from real footage.
Sora is built on a diffusion model, similar to GPT models, utilizing a transformer architecture. It represents videos and images as collections of data patches, akin to tokens in GPT, allowing for training on a diverse range of visual data. The model also employs techniques from DALL·E 3, enhancing its ability to follow text instructions accurately in generated videos.
For more information and to watch the examples, visit OpenAI’s website.
This evolution of AI is amazing. I look forward to the moment when you can really make full movies with just a few words. And the one minute of content is definitely a big step forwards. Other models are able to create between one and three seconds before they run completely amok.
What bothers me here is this “ethical use” limitation that is imposed on the tool. It should be up to the artist what they want to create with it. And no technical limit.
“What bothers me here is this “ethical use” limitation that is imposed on the tool. It should be up to the artist what they want to create with it. “
i think the ethical use problem isn’t directed at artists ( you’re right, they’re the ones who should decide ). But just imagine how powerful this is for fabricating fake news, propaganda, political and societal manipulation. You can bet a whole lot of groups besides artists are very interested in this.
You can trigger wars between countries with it, or civil wars….
Of course they are. But this adds a whole new dimension of censorship.
Yea i look forward when every single movie, game and artwork going to be same soulless trash too
Aren’t they already? Look at the stuff Netflix spits out regularly. Yes, we’ll have a LOT more of that. But then the truly creative work will just stand out more against the background of mediocrity. I feel like allowing smaller teams to work on bigger projects (through AI assistance in “labor-intensive” areas) will only benefit creativity.
This reminds me when Photoshop became a thing. I heard basically the same concerns back in the days. Because now everybody can edit images … ^^
I am the opposite of pessimistic here. The quality will rise. Because now even graphical untalented people can do an amazing job.
Without it setting a future precedent, in this case a 100% agree with Tiles 🙂