Painting AI said: Art creation is not just the exclusive product of human beings.
The artist said: Creativity is the last bastion that humans have.
In 2016, Google held a charity sale, and 29 artificial intelligence works created by the DeepDream painting system raised a total of 98,000 US dollars.
In 2018, "Portrait of Edmond Bellamy" created by artificial intelligence Obvious sold for $432,500 at Christie's.
Today, you can get the covers and illustrations of various commercial posters, publications, and content products through AI painting; you can also input a photo, adjust the corresponding parameters, and get a two-dimensional painting carefully drawn by AI for you, whether it is Overlord Luo Xiang, Hollywood star Leonardo DiCaprio, or the gray-haired KFC grandfather... AI can instantly satisfy all your fantasies.
From being controversial to being played badly, AI painting is being accepted by the younger generation.
So the question is, what is the logic behind the amazing AI painting?
Can we create art with AI?
Has the artist's fortress been taken over by AI?
In this regard, West China Metropolis Daily and cover news reporters interviewed Lu Jiancheng, Dean of the School of Computer Science (School of Software) of Sichuan University, Dean of the School of Intelligent Science and Technology of Sichuan University, Lu Jiancheng, as well as a number of people in the field of painting and Sichuan Yishang Law Firm. Lin Xiaoming.
Portraits drawn by AI
It once sold for a sky-high price of about 3 million yuan
Recently, the AI painting "Secondary Me" on QQ Small World has become popular on the Internet with its exquisite painting style.
How to output and transform the AI technology of two-dimensional handsome men and beautiful women is really amazing.
The reporter "played" for a while, and found that it can not only help the fat "Little Plum" Leonardo DiCaprio to return to the peak of his appearance in the "Girl Killer" period, but can even make the old man in the subway look at the emoji package on his mobile phone. Pulled to the same starting line as "Little Plum", this is completely different from the previous impression of AI paintings that do not output well and often "roll over".
In 2018, Christie's auctioned a portrait drawn by an AI program at a sky-high price of 432,500 US dollars (about 3 million yuan) in New York, USA.
The work, titled "Portrait of Edmond Bellamy," depicts a man wearing a black suit jacket with a white shirt in a hazy way.
The artificial intelligence that created "Portrait of Edmond Bellamy", called Obvious, was jointly designed by three 25-year-old youths from Paris, France.
The team collected 15,000 portraits and fed them into an algorithm called a GAN, which then learned the "rules" of those portraits and created new images based on those rules.
Behind the noisy image
hidden magic world
So here comes the question, what is the logic behind the amazing AI painting?
How to explain the recent popular diffusion model?
In this regard, Lu Jiancheng brought a first-hand interpretation.
There are many different methods of AI painting technology, but there are usually two types: one is style conversion technology; the other is image generation technology, including the recently popular AI painting technology based on diffusion model.
"The principle of AI painting is to learn from existing data sets through deep neural networks, so that the output painting results show an interesting feature. The focus of this process is on the organization and collection of data sets. This data set includes some samples. That is, the input data and the expected output. The deep neural network learns this mapping process from input to output, accompanied by some randomness, which produces a peculiar effect. The designer reflects the expected result in the data set, The AI model simulates the results expected by the designer as much as possible from the learning of the data set." Lu Jiancheng explained.
So, what is a diffusion model?
"The diffusion model has a more complicated mathematical definition, which is inspired by non-equilibrium thermodynamics. Intuitively, by adding noise, a clear image gradually becomes a noisy image with nothing. Is it possible to obtain a method from What about restoring the earliest clear image from the noisy image? The deep neural network model plays an important role in it. By learning the process of gradually changing from the clear image to the noisy image, the deep neural network model can learn to restore the original clarity from the noisy image The inverse process of the image. This is the so-called diffusion model. This is a surprising result, any noisy image can get a clear image. We can imagine what a magical world is hidden behind the noisy world."
However, after interacting with AI more, we will also find that "early" AI painters will also "roll over", such as "merging" the arm of "Little Plum" with other objects.
Why does AI "roll over"?
"The AI model needs to learn from the data set. This data set is designed by the designer, and it includes the input image and the expected output image. The learning process is that when an image is input to the AI model, the AI model outputs an image. The output The image is compared with our expected output. If it is not correct, we adjust the AI model parameters until the output image meets our requirements. It seems that the process of learning to draw is still very simple." Lu Jiancheng introduced.
However, when you feed the AI some data that it has not learned, the "rollover" happens." This is an interesting topic. The 'rollover' is because the output results are not consistent with the results we expect. Perhaps, if you put If it thinks it is an unexpected effect, then it is a new idea." Lu Jiancheng said, "The data set we design and learn is limited. When the input data is not learned by the AI model, or it is different from what it has learned before Things are too different, and the AI model has a certain degree of randomness in processing. However, sometimes it will produce some unexpected effects.”
So, does AI painting have "money scene"?
The answer is yes.
Lv Jiancheng said that AI paintings have been recognized as works of art, and the auction price of "Portrait of Edmond Bellamy" was about 3 million yuan is an example.
But the greater commercial value of AI painting is actually in the field of decoration.
It can be widely used in industrial design fields such as public space, private space, and daily necessities. Compared with the creation of traditional decorative patterns, AI generation has outstanding advantages.
As the main way of entertainment and social media, the AI painting tool itself also reflects its commercial value, such as Midjourney and so on.
In addition, as an auxiliary means of creation, the results generated by AI can also help artists and designers obtain creative ideas and make their works more imaginative. Therefore, AI painting has already been applied in many industries such as game design.
some artists worry
AI painting "steals" human artistic creation
Creativity is the last bastion that humans possess.
Eric, the lead game graphics designer of the game "Dead by Daylight", is one of the artists who resisted AI painting.
On social media, he shared several sets of images of Disney, Nintendo and other well-known IP characters made with AI paintings, and said that they were all drawn by AI tools.
Eric’s purpose in posting these posts is to show that AI painting “steals” human artistic creation.
The works created by artists who spend a lot of energy are trained by AI as data, and the generated paintings are more like copying or splicing.
Illustrator Bunning’s point of view is similar to Eric’s. Bunning said that he will not use AI painting programs or forward related pictures for the time being. “As a practitioner in the field of painting, I don’t encourage this kind of painting style splicing behavior. If In the future, AI painting will be of better use to me, or it can help me and make my creation smoother, so I will not reject AI painting.”
"Some friends in the game industry around me use AI as a tool to generate different sketches, and you can also see on Weibo that they use AI paintings to refine their works." Artist LKiKAi_ said.
However, she also said that AI will indeed have a negative impact on the painting industry, "It will affect some freelance painters, such as those who turn real people into hand-painted painters, and some painters who mainly draw avatars. After all, the pictures generated by AI are likely to be fake. Some of them can still be seen, and these painters will indeed lose some potential customers.”
Regarding the current problems of AI painting, Bunning said: "Now we have no way to stop the development of AI painting, but we can keep our own point of view. It is undoubtedly the progress of human technology, forming a very intelligent way of painting, which is convenient. It has helped some people, but at the same time there are problems, it lacks the author's feelings and creativity. Therefore, artists who are truly creative and have ideas don't have to worry about whether AI painting will affect them."
intellectual property disputes
Who does the copyright of AI paintings belong to?
"AI painting can allow people to express their individuality more conveniently, and allow novice painters to accidentally create master-level works. It is a very interesting thing to add fun to our lives." Lu Jiancheng said to the future AI The outlook for painting is optimistic.
So the question is, who is the owner of the AI painting?
Caselles-Dupré, one of the creators of the "Portrait of Edmond Bellamy" team, said: "If the author is the person who creates the image, then the owner of the AI painting is the machine. If the author is the person who has the idea and decides to implement it, then The owners will be us."
"Of course, when using images generated by AI, attention should be paid to intellectual property issues. At present, the copyright ownership of AI paintings is still controversial in the industry. Therefore, it is generally not recommended to use AI when the copyright ownership of the results generated by the platform is not clear. Painting for commercial purposes." Lu Jiancheng suggested.
Lin Xiaoming, a partner of Sichuan Yishang Law Firm, said, "According to the Civil Code, natural persons have the right to portrait, and have the right to make, use, publish or permit others to use their own portraits according to law. Under certain circumstances, the act of putting photos or emoticons of others, including celebrities, into AI programs to turn them into cartoons may be suspected of infringing on the portrait rights of others, but specific analysis needs to be carried out according to specific circumstances.”
Lin Xiaoming said: "According to the law, a portrait is an identifiable external image of a specific natural person reflected on a certain carrier through video, sculpture, painting, etc. With identifiable external image characteristics, if it can be identified by ordinary people as a specific natural person, there is a high probability that it will constitute infringement, otherwise it will hardly constitute infringement.”
Although the development of artificial intelligence has just started, the goal of general intelligence is still far away.
However, the current progress of artificial intelligence is already changing our lives and empowering our production.
Lv Jiancheng said frankly, "I believe that artificial intelligence can be seen in all aspects of life and production, such as medicine, industry, agriculture, education, home furnishing, cultural creation, commerce, tourism and other fields, and intelligent technology will be fully applied. "
In the future, talents in the field of artificial intelligence will go deep into positions such as algorithm engineers, program development engineers, artificial intelligence operation and maintenance engineers, intelligent robot R&D engineers, and AI hardware experts.
Lv Jiancheng said that as a national first-class undergraduate major construction site, Sichuan University's artificial intelligence major has covered undergraduates to masters and doctoral students.
The courses cover four major categories of AI basic courses, AI core courses, AI applications, and AI practice, especially in the interdisciplinary fields of basic theoretical research on artificial intelligence, intelligent image processing, speech recognition, natural language processing, industrial intelligence, and smart medical care. Deep accumulation.
So the question is, artificial intelligence is changing our lives, what should humans do?