Express the image in your mind in one sentence, as detailed as possible, then click, and within seconds you have a beautiful, highly detailed photo. Of course, you can also just enter a few vague words, but the more vague the description, the more surprising the results. Even if you don’t even have a brush, you can still “draw” a picture “A night full of stars” like Van Gogh’s or paintings “Impressive, sunrise” by Monet. If you’re more imaginative, you can also “draw” a cyberpunk-style onion, or a Cthulhu horror-style Chinese ink landscape.
The threshold for digital art has never been this low, until humans invented AI (artificial intelligence) painting. This year, some AI painting support systems have become the hottest topic in the art world and quickly spread to a large number of ordinary users. The AI seems to be violently subverting the art, in the same way it has disrupted the game of Go.
Just as the AlphaGo system learned the profound meaning of the mind game by repeating Go games over and over again, the painting AI has also absorbed the work created by a large number of artists. artist, thereby making it easy for users to create any drawing style they want. And this has worried many artists. Obviously, learning to draw for a lifetime, but now it’s not as good as people typing a few words?
Worrying even more are those who rely on digital drawing and design for a living, such as in games, film, industrial design and other industries. As they see the rapid development of AI painting systems, a gloomy future called “unemployment” seems to be approaching.
But the rapid development of AI painting has brought with it a series of ethical problems. When an artist’s work is randomly captured by AI and learned from the Internet, that is, every time they create a new work, they are training themselves an invisible and powerful opponent. Many artists have protested, saying that AI should be banned from using their works, but the developers of this software are busy upgrading their systems, so they don’t have time. feedback.
Will art die?
In April of this year, Disco Diffusion, an AI painting program based on Google’s engineering platform, suddenly became popular. Disco Diffusion’s interface is quite unfriendly and almost impossible to get used to by people without programming experience. Fortunately, some netizens have written up a detailed tutorial, changing the parameters to the right position, so that after entering the command in English and clicking to run, after only a few dozen minutes, you finished work can be obtained.
After that, Open AI organization released a promotional video about its AI painting system called DALL-E2. This video surprised many people that the AI can draw pictures to such a high level, and it is also much easier to manipulate than Disco Diffusion.
“Unbelievable”, many digital artists and tech bloggers have to say. “I can’t imagine the existence of this.”
Compared with Disco Diffusion, the accuracy and clarity of DALL-E2 are significantly improved. Over the next three or four months, the big Internet technology companies jumped into the game. Google released Imagen and Parti, Facebook released Make-A-Scene, Microsoft released NUWA and a startup company released Midjourney… In China, Baidu also launched an AI painting tool called Wenxin Yige .
These systems have their own characteristics. Some people argue that Disco Diffusion’s work is often beyond the imagination of users, sometimes surprising because it seems to harbor an artistic ambition of its own, while Midjourney’s work is close to the imagination. more people, can more accurately translate the image in people’s mind.
AI painting was quickly promoted spontaneously at an unprecedented rate. Midjourney already has 1.5 million users and is expected to increase 5 times within a month. And users themselves are constantly improving this technology in the process of enjoying their leap from rookie to artist.
What is even more surprising is that these systems show progress at lightning speed. Last month, its skills in drawing human faces might have been crude, but now you can hardly tell the work of an AI from an artist with years of experience. Some newer forms of application have also been developed rapidly, such as Dall-E which can make characters in an AI comic book consistent and can be used in comics and animation. . The Clip feature built by Open AI can directly use a series of pictures to make animations. The short film “The Crow” created by Clip won an award at this year’s Cannes Short Film Festival.
And in August, the battle over AI painting technology between the big giants entered a new chapter with the release of Stable Diffusion. Not only possessing better drawing capabilities based on a database of 5.9 billion images, but also because it is an open source software. Global technology developers have been able to develop many new tools based on their models, and AI paintings have created an effect like a nuclear explosion. For example, in the first week after the open source was released, a plug-in for Photoshop based on Stable Diffusion appeared, easily bringing Photoshop into the era of AI paintings.
But also from this time, the controversy about AI-drawn pictures has spread from the Internet to real life. In the “Digital Art” competition at the Colorado state fair in the US, a contestant named Jason Allen submitted an AI-powered painting called “Théâtre D’opéra Spatial” and won the top prize. There was a lot of excitement and support, but other contestants and some netizens accused the guy of cheating. But the problem is that Allen did not break any rules, the work is also clearly noted as “Jason Allen uses the Midjourney tool.” Furthermore, the contest organizers have not said that they will withdraw the prize. reward.
“Art is dead”, In the post-contest interview, faced with heated controversies, the 39-year-old answered firmly: “AI wins, humans lose”.
Will artists lose their jobs?
Compared with AI systems capable of playing chess and writing poetry, painting with AI brings a sense of deeper impact on people. Not only does it announce that machines are reaching the pinnacle of artistic creation, but it also seems to be able to capture more of an aesthetic that humans pride themselves on. Besides, it is also a real threat to some professions. For example, in game graphics, film and television art, industrial design and other fields, AI seems to be able to replace artists with a little training and sometimes its imagination. even beyond humans. On the Internet, one of the most discussed questions about painting AI systems is: “Will they put artists out of work?” or “Is game art killed by AI?”…
“Artists can draw 100 types of piles of bricks without repetition, but the AI can randomly generate 10,000 types for you to choose gradually.” an artist working in the game field said. She said her team used Stable Diffusion to create the icons and wreckage images in the game. This is a job that used to require a lot of time and manpower, but can now be simplified to just one person to choose an image. She also uses AI to design in-game medals, by entering keywords and hanging up overnight. The next morning, the result will be thousands of products. Most of them can be put to use immediately, she said, while manual workers can only draw about a dozen or so a day.
And one of the biggest headaches in the game industry is the communication barrier between the producer and the art production staff. Liang Qiwei, founder of gaming company Lingyoufang, the production unit of the Shadow Blade game series, says that producers are often unable to clearly express their wishes to art creatives. , making them only able to find out on their own. This leads to the production process being repeated many times, which is costly and time consuming. But with AI painting systems, they will bring huge changes. In the initial design phase, AI can generate a large number of sketches based on the settings of atmosphere, light, style, texture, etc. On this basis, the producer and the designers artists can quickly understand each other’s needs.
In China, the game industry is the leading sector Experiment with painting with AI first. The industry has extremely high demands on the quantity and efficiency of artistic resources. However, there is still no effective method for how to “tame” this monster of limitless ability.
Liang Qiwei has been thinking about this recently, and he thinks these AI platforms will make more sense for the initiators, than for the performer of the art. In other words, how to draw with AI is only secondary, more importantly, AI will tell us “what to draw”. For example, before setting up a project, a producer can hand over the ideas he has come up with to the AI, add some keywords and test iteratively to see what can be created. and whether it matches the “feel” that I want. And AI will work tirelessly to meet any of these absurd requests, solving a process that art creators often consider a “nightmare.”
“Join if you can’t beat it”, Simon_Awen, a designer and tech blogger, was optimistic. “AI won’t make me unemployed, it will help me get out of work sooner.”
At first, when he first tried painting with AI, he found that the technology was quite rudimentary, could help create textures but required him to add details. However, in just a few months, AI painting has made great leaps and bounds. Simon feels that AI has now reached the level of mid-level artists and this will inevitably lead to a group of artists being displaced. But this is only a few months of launch and no one can imagine how far it could reach in the next few years. He realized that if he used this tool, the cost of developing art projects would be reduced by at least half.
And in fact, some artists will not only not be eliminated, but will become “invincible” because they now have an unprecedented tool. When artists are familiar with how to use and command AI painting, the works created will be much more amazing than the paintings today. Because in the hands of an ordinary person, AI often loses control, but in the hands of artists, they are like tamed steeds. Simon_Awen says: “You can be a very good AI-powered artist if you want to.”
Make art or steal art?
When people enter the command: “Little girl, big head, Nara Yoshitomo”, the AI will draw a picture similar to the classic style of famous artist Nara Yoshitomo in just three seconds. But, who is the author of this picture?
Is it Yoshitomo Nara, the user or the programmer who wrote the code for this AI system? When an artist’s lifetime work is accurately learned and imitated by the AI, how can that artist protect his or her rights?
Greg Rutkowski, a Polish fine art artist, is concerned. He was once known for his epic fantasy works in the style of oil painting, but these works have now become the subject of much parody by Stable Diffusion. When he searched his name on the Internet, all the paintings that came up were AI paintings and his own works were drowned. In just one month, his name was used by the AI more than 93,000 times as a keyword, while names like Michelangelo, Picasso and Leonardo were used just over 2,000 times. In a recent interview, Rutkowski lamented that he felt his career was threatened.
For artists’ concern, David Holz, founder of the AI painting system Midjourney, explains that it is “a bit fanciful”.
Because according to Holz, many famous artists even want the AI to be better able to “steal” their style, so that it can be used as part of the creative process. A lot of artists also say that they find this tool very interesting, because it’s like an art student when it references an artist’s name to create a painting. Like how an art student creates something inspired by an artist.
But not everyone is comfortable with them. A group of Japanese cartoonists have voiced a collective complaint and asked the developers to strictly prohibit allowing AI to imitate their drawing style. Some international artists are even forming alliances, hoping to promote new policies and regulations to govern painting AIs. Two Berlin-based artists have set up a website where other artists can search to see if their work has been included in the AI’s training database. There are also some voices calling for artists to at least be paid, if not able to stop the AI from copying themselves.
However, to the artists’ objections, the AI developers either responded perfunctorily, or simply remained silent.
Digital artist RJ Palmer argued with an AI software developer on social media, but was later “blocked” by this person. Palmer later publicly stated that it is the responsibility of AI developers to ethically collect images to train their AI, not to constantly steal the brains of artists.
Freelance artist Andy Bayo describes the painting AI as a Pandora’s box. He summed up his three arguments: Is it ethical to train an AI with a large number of copyrighted, unlicensed creations? Is it ethical to allow people to create new work in the style of photographers, illustrators, and designers without compensation? Is it ethical to charge for services based on someone else’s work?
Obviously, he can only discuss this issue in the area of ethics, because the law is not yet able to regulate this new thing.
But equally disturbing is that the AI-generated pictures also show some dangerous tendencies. When freely created and published, users can create disturbingly violent and sexually explicit images. Midjourney founder David Holz himself acknowledged the existence of this danger in an interview with the media. He said that every time the operation team sees such photos, they will clean up and place a ban on sensitive words when necessary. But Midjourney’s engineering team is only ten people, and the photos it creates are growing exponentially every minute.
Another danger is that they can create images from real faces and transform them. Stable Diffusion allows users to create celebrity portraits, even nudes. Even a stable Diffusion-based pornographic site has appeared.
After the Queen’s death, many users created pictures of the Queen in various styles to express their condolences, but there were also some who created controversial images. .
Since Stable Diffusion is open source, the engineering team cannot manage everything, so it is difficult to restrict harmful information based on this model. Although it has been condemned many times, but in the context that everything has just begun like today, the competitive advantage brought by this open spirit makes the development team quite shy in tightening the management system. physical.
Are AI paintings art?
Over the past decade, artificial intelligence technology has made great strides thanks to three important factors: Graphics processing chip (GPU), deep learning function (Deep Learning) and big data (Big data). Data).
During the 2010s, the performance of GPUs developed by companies like Nvidia for video games increased at an alarming rate. Its ability to handle large amounts of computation has made it ideal for use in neural networks – programs that mimic the way cells in the human brain interact. This learning behavior provides the foundational competencies for AI. The explosion of big data in various fields on the Internet – photos, paintings, music, Go … has provided endless resources, making it easy for AI to capture the intellectual achievements of human beings. accumulator.
The most basic technology of the painting AI is the same as that of AlphaGo, an AI that is good at Go, both of which allow the system to dig deep into the works of people, thereby creating actions. vi imitation. Just like the debate that accompanies AIphaGo, the strongest suspicion and disdain for AI paintings comes from doubts about its uniqueness. In other words, AI lacks the so-called “human taste” and can therefore only produce “fake”.
In this view, AI cannot express the subtle human emotions, as it cannot depict the faint smile of the Mona Lisa. Even if some touching details are generated by AI, the appeal of art will be lost because we already know that the detail comes from an algorithm, not the inner expression of the person. living.
However, with AI’s ability to learn, can this problem one day be solved?handle?
Zhang Ji, an associate professor at the School of Arts and Humanities of the Chinese Academy of Arts, believes this is indeed a false statement. “I think it is not only human, but also better than many people.” she confirmed. “Because the ability and learning speed of AI is much higher than that of ordinary people, it will draw better than ordinary art people.”
Zhang Ji has been following the art of artificial intelligence for the past seven or eight years. In 2016, when Google’s Deep Dream held its first artificial intelligence art exhibition, and in 2018, Christie’s auctioned off “Edmond de Belamy” created by a French engineering team, she felt those works all have flaws, but the market recognizes them. Currently, the quality of AI painting and animation art has taken a big step forward.
Midjourney founder David Holz maintains a modesty when it comes to the art of AI. He believes that art is often about stories and emotions, but AI is not about stories and emotions.
Since ancient times, art has been a unique human ability, but after the emergence of an unprecedented human superpower like AI, can the works created by it be called art? Will AI change the definition of art?
As AI advances by leaps and bounds, this philosophical question will likely persist for years. One of the most discussed perspectives is that the work of art has more to do with the author or the viewer. According to some views, the meaning of a work of art is constantly recreated in its interactions with readers, spectators, and viewers. And art and artistic concepts never stop evolving. AI is seen as a new art form, and art forms will also shape the concept of fine art.
For more than 100 years, the old art style has been under constant threat from new technologies. For example, photography, film… but the art of painting still exists, although some professions such as portraiture have almost died out. The advent of drawing software like Photoshop has also caused another shock, but the art still exists. So this time, will things be different? Let time give the answer.
Refer Sina, iFeng