Enter a paragraph and "draw" a picture - the painting skills of artificial intelligence, attracting many professional painters and zero basic users to taste. The essence of AI painting is calculation. After receiving the instruction of "language description", it can restore the image according to its own understanding. In the future, when AI technology is applied to art creation and other fields, attention should be paid to preventing potential risks, so that technological progress can better benefit society. The computer can automatically parse and generate corresponding paintings by inputting a descriptive text without using brushes and pigments. At the 2022 World AI Conference, the exhibition of AI painting amazed the audience. AI has also begun to dabble in some fields that used to be exclusive to human creation, such as painting, calligraphy, writing and composition. How does AI paint? What kind of sparks will come from the current encounter between technology and art? Are there any problems in content and copyright? From text to image, the essence of AI painting is computing AI painting is a generation process from text to image. Input a paragraph of words to generate a picture, and the essence is computing. Briefly, the computer can recognize the relationship between specific image elements and text through a lot of learning. Similarly, after receiving the "language description" instruction, the AI program can restore the image according to its own algorithm. The idea of setting up computer programs for drawing has been around for a long time. As early as the 1970s, some artists developed a computer program for operating the mechanical arm to draw on the drawing paper according to instructions. In recent years, artificial intelligence technology is changing rapidly, and researchers try to design computer programs for automatic drawing. However, for a long time in the past, the works drawn by artificial intelligence are generally not good enough. They are often just the combination of some fuzzy image elements, and cannot be called a complete painting. Since this year, AI painting technology has rapidly "evolved". When it comes to the reasons for technical breakthrough, Xiao Xinyan, chief architect of Baidu Wenxinyige, believes that this is the result of the rise of pre training big models, big data training and the emergence of diffusion models. Specifically, the pre training large model enhances the universality of AI and becomes a new base for AI technology and application; In the training of big data, through parallel learning among many high-performance GPU (graphics processor) computing resources, the computer can complete a large amount of data learning in a short time. In recent years, almost all the technological development of AI has benefited from the progress in these two aspects. For AI painting, the emergence of diffusion model is crucial. The principle of the diffusion model is to gradually add noise artificially to make the image gradually "fuzzy", and then continuously learn the noise removal process, so that AI can gradually restore a clear image from a picture that is completely noisy, that is, "draw" the image. "This process is similar to human learning. Generally, people learn painting from copying, and so do machines. The image it initially generated may be very fuzzy, but the computer will constantly revise, so that the output of images is becoming clearer and richer," Xiao Xinyan said. The diffusion model enables AI painting technology to leapfrog, which not only improves the painting quality rapidly, but also shortens the generation time to a few seconds. People