OpenAI unveils Sora, a text-to-video AI generator
OpenAI announced Sora, a text-to-video AI that generates high-resolution videos up to one minute long without audio. Unlike previous text-to-video models like Runway Gen-2, Sora achieves superior internal coherence and adherence to physical reality and cultural references. Examples include a cartoon kangaroo dancing in a disco and photorealistic golden retriever puppies playing in snow. However, imperfections remain, such as a panda with only two legs. OpenAI operates opaquely regarding training data, possibly using synthetic data from Unreal Engine 5. The primary concern is the potential for rapid, large-scale production of fake news videos. Italian cartoonist Daniele Giardini highlighted a risk to the online economy based on cat videos, as AI-generated content may undermine authenticity. Sora is currently available on a limited basis.
Key facts
- OpenAI announced Sora, a text-to-video AI.
- Sora generates videos up to one minute long without audio.
- Sora achieves higher coherence than previous models like Runway Gen-2.
- Examples include a cartoon kangaroo dancing and golden retriever puppies in snow.
- Imperfections include a panda with only two legs.
- OpenAI has not disclosed Sora's training data; may use synthetic data from Unreal Engine 5.
- Risk of rapid fake news video production is a major concern.
- Daniele Giardini warned Sora could disrupt the online cat video economy.
Entities
Artists
- Daniele Giardini
Institutions
- OpenAI
- Runway