OpenAI, owner of ChatGPT, launches a model that creates realistic videos with artificial intelligence from texts

[ad_1]

Sora can create videos of up to 60 seconds from brief descriptions, according to the company. For now, only visual artists have received access to the tool, to say how it can be useful for their creations. Meet Sora, a generator of realistic videos based on artificial intelligence, made by the owner of ChatGPT OpenAI, creator of the ChatGPT robot, revealed this Thursday (15) an artificial intelligence model that creates realistic videos from short text. Named Sora, it was released for expert analysis and is not yet available to the public. “Sora can create videos of up to 60 seconds with highly detailed scenes, complex camera movements, and multiple characters with vibrant emotions,” explains OpenAI. “The model understands not only what the user asked for in the prompt [comando], but also how these things exist in the physical world.” Video created with Sora artificial intelligence shows mammoths walking on ice Disclosure/OpenAI Demonstrations of Sora creations include both animations and realistic-style images (see examples). No videos were recorded by humans, but generated by artificial intelligence from short texts, according to OpenAI. For example, the video showing a group of mammoths walking on ice was created with the following description: “Several giant woolly mammoths approach walking across a meadow snowy, their long woolly fur blows lightly in the wind as they walk, snow-covered trees and dramatic snow-capped mountains in the distance, mid-afternoon light with thin clouds and a high sun in the distance creates a warm glow, the camera’s low view is impressive, capturing the large furry mammal with beautiful photography and depth of field.” Initial plugin text To reduce the risk of misuse, OpenAI promises to adopt security measures before making Sora available in its products. This includes working with experts in areas such as misinformation and hateful and biased content. Furthermore, a group of visual artists, made up of professionals such as designers and filmmakers, received access to the tool to give their contributions on what can be done to make it useful for their creations. “Sora serves as the foundation for models that can understand and simulate the real world, a capability that we believe will be an important milestone in achieving AGI [Inteligência Artificial Geral]”, says OpenAI. How Sora works Video created with artificial intelligence Sora shows character animation playing with a candle Disclosure/OpenAI Sora uses a technique called “diffusion”, which creates images from random points. At the beginning of the process , the video has an appearance of static noise, the effect of old TVs that have no signal. And, little by little, the visual is transformed into something that can be recognized by a human being. The technique is similar to that of robots that can create photos from user descriptions. One of them is DALL-E, which was also created by OpenAI and served as the basis for the development of Sora.”[O Sora] uses the DALL-E 3 (robot) recapture technique, which involves generating highly descriptive captions for the visual training data. As a result, the model [Sora] is able to more faithfully follow the user’s text instructions in the generated video,” says OpenAI. Not everything is perfect. The company claims that, in addition to creating videos from scratch, its new artificial intelligence model is capable of creating continuations of existing videos and generating videos from still images. And admits that it also has “weaknesses”. According to OpenAI, Sora may have difficulty reproducing the physics of a complex scene and may not understand cause and effect logic. “Por For example, a person can take a bite of a cookie, but then the cookie has no bite mark”, he explains. “The model can also confuse spatial details of a ‘prompt’, for example, mixing right and left, and can have difficulties with accurate descriptions of events that occur over time, such as following a specific camera trajectory.” Security measures OpenAI promises to take several security measures before making Sora available in its products. It says that in the testing phase , the video generating tool is being challenged by experts in areas such as misinformation, hateful content and bias. “We are also creating features that help detect misleading content,” says the company, which promises a detector for videos created by Sora. According to OpenAI, the presenting text for ordering a video will be checked to identify commands that lead to a violation of the company’s usage policy, such as those that require extreme violence, sexual content, similarity to celebrities, etc. “We cannot predict all the beneficial uses people will have for our technology, nor all the ways people will abuse it,” says OpenAI. “That’s why we believe that learning from real-world use is a critical component to creating and launching increasingly secure artificial intelligence systems over time.” How robots that create photos work How robots work that create new images in seconds

[ad_2]

Source link