The organization OpenAI is revolutionizing the field of AI (artificial intelligence), by presenting the model GPT-3 - ChatGPT that can generate text in a human-like manner, and DALL-E - a system for generating images from text. Now they have introduced Sora - a model for creating videos from text.
At OpenAI, they claim to be teaching AI to mimic movement in the physical world, with the goal of training models that help people solve problems that require interaction with the real world. Sora can generate videos up to a minute long based on user instructions, including complex scenes with multiple characters, specific movements, detailed background objects, and even different camera angles. The model not only understands what the user asks of it, but also how people or objects are supposed to interact in the real world.
The model is not yet perfect, as it may not understand the consequences of a specific action, for example, in a video where a person takes a bite of a cookie, the cookie will remain whole. Sora also might get confused between right and left or in scenes that span over time.
The Sora model is not available (at least for now) to the general public. OpenAI is offering it to a group of researchers who will assess the existing risks in it. In addition, it has also reached out to a group of designers and filmmakers to receive feedback that will help advance the model, so that it will be better for creators. According to OpenAI, the reason they exposed Sora at this initial stage is because they want feedback from outside the organization and also for the public to see the future of the AI field.
The videos released by OpenAI are very impressive, they include aerial footage of a house by the sea, an astronaut walking in space, elephants walking in the snow, and a woman walking at night on a rainy street. Click here for more articles on The Gadget Reviews website.