Text / Yangcheng Evening News reporter Chen Xiaonan
From Disco Diffusion to DALL·E, Awen's heart was shaken twice.
In 2015, Awen, a 25-year-old data analyst, entered the visual design industry.
Since then, he has used auxiliary tools that can perform style transfer, allowing the image to be processed to change various styles.
And by entering a few commands, he can draw a picture-a kind of thing he never dared to think about before.
Until the birth of AI painting tools one after another, the early adopters continued to subvert the designer's cognition again and again.
AI painting shakes the careers of animators and designers?
In April this year, at twelve o'clock at night, Awen was lying on the bed and swiping Weibo.
When he saw an artist friend post a painting that he claimed to be directly generated by AI-a concept scene map of cyberpunk style, he jumped up instantly.
Half shocked, the other half suspicious, he immediately went online to find the tool that his friend said - Disco Diffusion.
After studying the tutorial that night, Awen used Disco Diffusion to draw a sunflower, and all this required was to enter one sentence in Disco Diffusion: a sea of sunflowers under the stars.
There is a feeling of excitement and trembling.
He immediately sent the painting to several creator friends.
One stone stirred up a thousand waves, and Awen's circle exploded. Everyone was talking about it and eager to try.
On April 30, Arwen asked Disco Diffusion to imitate a painting by the 19th-century painter Henri Fontaine-Latour, but the resulting effect was less than one-tenth of the original.
He wrote on his social account: I am not delusional that AI can match or even surpass these artists.
On the contrary, all the test results are proving the greatness of these artists.
"But I was slapped in the face 3 months later." Awen laughed.
Since the birth of the first popular AI painting tool, out of curiosity and professional needs, Awen has been tracking the dynamics of related fields, and even joined a lot of AI tools in the internal testing group, witnessing AI painting all the way. growing up.
Until Stable Diffusion came out, AI painting has been able to highly imitate the artist's style.
The trend of AI painting is like a spring breeze, awakening many sleeping people in the cultural industry, and quickly attracting attention.
There are many voices on social networks. There are many voices worried about AI infringing copyright, trampling on creativity, and replacing painters. There are also optimists who have expectations for AI.
Is AI painting really shaking up the professions of animators and designers?
Having been immersed in the design industry for several years, Awen, who should have a sense of professional crisis, is the optimistic one: "At first glance, it is very efficient, but when it is actually used in actual work, it is not so brilliant. It’s my greatest experience.” He feels that the AI drawing tool is now in an awkward position, and it is too early for it to completely replace the painter, but it is a bit wasteful to only use it as an efficient tool that can mass-produce simple graphics .
As for whether AI painting can enter the cultural industry, "at least my peers and the company's decision makers are waiting to see." Awen said.
Soulframe is an indie game producer.
He believes that the current stage of AI painting may be more meaningful to the art demand side.
In other words, AI can help Party A become a better decision maker.
In his opinion, images generated based on demand-side prompts (prompts) are always more intuitive than a bunch of descriptive words, and can even awaken the imagination of "conservative" decision makers.
According to the industry report of the qubit think tank on AI-generated content (AI-generated content, hereinafter referred to as AIGC), in general, AIGC currently has a relatively limited overall impact on the content consumption field.
However, with the gradual deepening of the commercial application of technology, AI painting will have an impact on the talent structure of the content industry.
Since basic skills such as pre-rendering and overall color grading will be completed by AI, the focus of designers' work will shift to putting forward prompts—that is, how to clearly describe artistic ideas to AI, and improve their creativity, aesthetics, and storytelling and search capabilities.
"I think it is very likely that prompts will appear in the future." Awen has participated in the internal testing of many AI painting tools.
In the group, he can always see some people who are particularly good at writing prompts and can debug the most wonderful paintings.
An AI painting sells for $430,000
As AIGC matures, will AI become the cyberflow worker of the content factory?
Will the hands and minds of cultural and creative practitioners be liberated?
When DALL·E came out, Awen tried to use it to explore and solve the classic problem of "difficult party A": can the elephant turn around.
DALL·E easily drew two pictures showing the back and front of the same elephant.
Therefore, many people jokingly call it the "Terminator of Party A".
Painters and designers suffer from the changing requirements of Party A, which is a tacit problem for the creative industry.
After immersing in the existing AI painting tools on the market, Awen felt that the so-called "Party A Terminator" is still just a "spit" style joke.
"In actual work, the details that need to be adjusted are more efficient if you have to do it yourself."
In Awen's view, only when AI can get rid of imitation of others and can better handle details can it be a truly perfect design assistant and can truly liberate its hands.
"Of course, maybe it can replace me at that time." It's just that the sense of accomplishment of using AI to create works is much less than the joy of designing and creating by yourself.
"Sometimes I feel it is like a living person, like a hard-working Party B." Awen said.
In August, he released an AI painting imitating the style of Kawase Ba Shui, which attracted enthusiastic attention from netizens. He signed Stable Diffusion as a designer and himself as a design assistant.
Some netizens described that letting AI paint is like "casting magic", and the more "spells" you recite, the better the AI can understand the user's intentions.
"The next development trend is to improve the controllable generation of AI painting, that is, how to make the work more suitable for the user's instructions." said Ding Ming, a doctoral student in the Department of Computer Science of Tsinghua University and the author of CogVideo, an AI-generated video tool.
In the face of AI with a high degree of imitation, how should artists handle themselves?
Artist Qiu Zhijie once defined GAN (one of the three major models of AI-generated images) as the hottest emerging artist in the art world in 2019.
The French art organization Obvious used GAN to create 11 works of art, one of which appeared at a Christie's auction in October 2018 and was sold for $430,000, causing a sensation.
In September of this year, the "Space Opera" generated by the AI drawing tool Midjourney won the first prize in the digital category at the exposition art competition held in Colorado, USA, and the controversy continued.
Qiu Zhijie believes that the development of AI may force the "evolution" of artists.
Emotions, Consciousness, and the Future of AI
Seeing that their own works are included in the AI training data set, and the AI creates works that are exactly the same as their personal style, some artists express concern and anger.
According to CNN, a pair of artists in Germany and two collaborators have launched "Have I Been Trained", a search engine that can help artists find the LAION dataset to see if their work has been used for training.
Feng Yuan, a professor at the School of Communication and Design of Sun Yat-Sen University, once said that artificial intelligence will face a "spiritual problem": the human mind has emotional characteristics, and artistic creation is inseparable from such emotional characteristics and endocrine neurochemical mechanisms.
In the future, if AI can surpass imitation, evolve original capabilities, and even create with "consciousness" and "emotion", how should we reassess the value of AI art?
People always want AI to "be human".
In the sci-fi film "Blade Runner", the android Roy is full of poetry and emotion when he is about to die, and recites: The things I have seen, you humans absolutely cannot believe.
I watched the battleship burn next to the Orion constellation, and I watched the C-rays shine in the darkness near Tannweiser's Gate.
All these moments will be lost in time, just like tears in the rain.
"In fact, AI can more or less understand human emotions now." Ding Ming said.
Through mapping, when the operator inputs "a happy person", the AI can map "happy" and "smile" to draw a smiling person.
However, in practical applications, AI's understanding and simulation of emotions are still at a shallow level.
"As for whether AI can be conscious, there are many explorations in the research community, but most of them are just hypotheses." Ding Ming said.
At the end of the last century, American psychologist Bernard Balls and others put forward the consciousness model "Global Workspace Theory (GWT)", which is one of the main contemporary neuroscience consciousness theories.
It has been suggested that the current deep learning can already be based on GWT, combining the neural networks that deal with different modal conversions, that is, functional modules, into a system, thus moving towards the next stage of realizing general artificial intelligence.
How high the future of AI art can reach also seems to depend on the application process of the consciousness model.
The value of AI paintings is also related to the definition of artistic value and functions.
Qiu Zhijie said that the value of art is not only aesthetic, but also social, and it is a matter of political economy.
As for whether AI can create a new chapter in art history and even human beings, it needs to be verified by time and technology.