When the “barbarian” AI painting knocks on the door, where will the content track go?

Economic Observer reporter Ren Xiaoning reported in Beijing

After mid-October, Zhang Shiying felt that there was a warm AI painting trend in China. On her platform, the number of users increased by 10 times compared with the previous month. In the circle of friends, Xiaohongshu, and Weibo, there are people everywhere who are eager to try AI painting. Zhang Shiying is the first founder of the AI ​​creation platform YUAN. She told the Economic Observer that this is a scene she has never seen in the past two years in the AIGC field. “It’s too popular.” Before receiving the call from reporters, Zhuang Minghao, vice president of Quwan Network, had just finished a discussion on AI painting, and this topic was the topic that his friends were most keen to discuss recently. In the second half of this year, the topic of AI painting began to break from the niche circle to the public, and this speed far exceeded his expectations.

When the artist Xiaotao first came into contact with AI painting in August, she was a little dissatisfied. In October, the high quality of the pictures drawn by AI shocked her. There is a joke on the Internet. Some people ask why AI learns to draw so fast? The best answer is: because you give it learning materials, it really learns. This has also sparked dissatisfaction among some painters, who believe that AI is plagiarizing and plagiarizing the works of human painters, and announced a boycott. Under the hustle and bustle, AI painting has become a new outlet, foreign companies such as Google, Microsoft, Meta, and domestic Baidu have all come to an end, and entrepreneurial unicorns with a valuation of 1.5 billion US dollars have also been born. But the question is, how big is the trend of AI painting, will it be a new era that is about to change the process of human creation? Or will it quickly dissipate or become an auxiliary tool after the explosion?

wind from overseas

Although it only became hot in October in China, the wind of AI painting has actually been blowing overseas in July and August.

Zhang Shiying’s AI creation platform also has an overseas version, called imgcreator. The main users are from the United States, Brazil, Germany, Russia, Indonesia and other countries. An American novelist recently used imgcreator to create illustrations. He didn’t know how to draw before and needed a professional illustrator to match the illustrations. On the imgcreator platform, he entered keywords, and the AI ​​automatically generated pictures. He finally created a children’s comic book by himself, and it sold very well.

“For this novelist, AI painting is a tool to make money. He used to create a book in 2 to 3 months, but now he can create a book in a few days. There are more opportunities for trial and error, and the efficiency of making money will be higher.” Zhang Shiying tell reporters.

Overseas AI creation has become a dominant outlet, and the types of creation are not limited to paintings. The novels of the above-mentioned American novelists are also created with AI, using a platform called Jasper.ai. On this platform, users can quickly generate a short story with ups and downs of hundreds of words by entering a text summary of a dozen or so characters. In October, Jasper raised $125 million in Series A funding at a $15 valuation. The company has more than 70,000 customers and generated $40 million in revenue last year.

In addition to writing and painting, AI creation is being implemented in industries such as e-commerce, design, and marketing. “It is not only a painting, but something that can really help some people speed up in the industry and business.” Zhang Shiying said.

Also in October, Stability.AI, a British AI creation company, also received $100 million in financing, with a post-investment valuation of $1 billion. OpenAI, the AI ​​company behind DALL·E, the originator of the AI ​​creation craze, is reportedly negotiating with Microsoft and wants to get another round of financing. Previously, OpenAI was valued at nearly $20 billion.

Enthusiasm for AI in investment circles is returning. “In a big way, now everyone has finally seen the possibility of using AI, and an ordinary person can think of how to use it.” Tell the reporter that the investment circle has been questioning AI in the past few years because it cannot be implemented. Now, AIGC’s imagination is too big. He gave an example. There is a scene in “Westworld”. The game producer is talking while doing the game plot and completing the construction of the entire game scene. What kind of palace and what kind of castle does he need? What character appears, just need to speak to complete. At the time, I thought it was a sci-fi film. According to the current development speed of AIGC, it is very likely that it will become a reality.

“It has changed from a toy for the extreme customer group to an interesting thing for the general public.” Zhuang Minghao feels that the popularity of AI painting shows that AIGC has broken the circle.

Simon_Awen, a Weibo V and a partner of Ableslide, a PPT customization company, has been paying attention to AI painting. He wrote a long article in June to popularize the knowledge of AI painting. When it was updated again at the end of August, he almost rewrote a new one. The article, “It’s so sighing, I feel that AI painting has fast-forwarded to the battle of the gods.” At the beginning of writing the article in June, the AI ​​painting product was still in the internal testing stage, and users needed internal testing qualifications to experience it. In August, the big foreign manufacturers came to an end, and each launched more than one product and began to roll in. In August, Sta-blediffusion announced that it was open source and open to the public for free, setting off a carnival of AI painting abroad. Since its launch, more than 1.7 trillion generated images have been produced through its official platform DreamStudio alone. “It is foreseeable that there will be more and more exaggerated models and tools in the future.” Simon_Awen said. Regarding this phenomenon, someone commented on his Weibo, “Maybe looking back, this is the beginning of a new era.”

Technology is breaking through

AI painting is nothing new. As early as 10 years ago, Microsoft’s AI Xiaobing can draw, sing, and write poetry. In 2018, the Google Drawing applet caused heated discussions in the academic circle. But it was not until this year that AI painting attracted high attention in the public circle. The reason behind it is the technological breakthrough of AIGC in the model.

AI painting in 2022 can make a person who has never been able to paint quickly become a “magic pen Ma Liang”. The 7-year-old daughter of Zhang Lijin, a senior researcher at the Center for Big Data and Social Computing at Communication University of China, recently became fascinated with AI painting. The production scene is very simple. For example, if you want a picture of a jellyfish, you only need to open the applet and enter keywords such as “jellyfish, seawater, tentacles” in the text box, and you can generate various styles of jellyfish in a few seconds. If you are not satisfied, click to generate again and there are new pictures to choose from. The more keywords there are, the more imaginary the generated image will be. In the past, people had imagination, but lacked a pair of hands that could draw. Now people only need to describe with their mouths, and AI is the hands that can draw.

This painting method has lower threshold and higher quality. Also draw a picture of a cat. The cat drawn by AI in 2017 is composed of several triangles or rectangles. In 2022, after inputting a long string of words like magic spells, the AI ​​​​draws a delicate and fashionable cat with a steel and gold wire armor, piercing eyes and a majestic look. “In April this year, a paper about diff-fusion became popular in the industry, and there was a big breakthrough in the content quality of the entire AIGC.” Zhang Shiying told reporters that AI could also create content in the past, but the quality was not high. Commercially possible. This year, Google, Microsoft, and Meta all launched their own AIGC products. The second half of the year was especially lively. Each of them launched their own AIGC models under the diffusion framework and commercialized them in various fields.

The domestic Baidu company also launched the AI ​​painting platform “Wenxin Yige” in August this year. The relevant person in charge of Wenxin Yige told reporters that the initial AI painting was to completely imitate the way of human painting, the cost was very high and only relatively abstract results could be obtained. In 2022, there will be a technological explosion in AI painting, and a diffusion generation model will be introduced. After this type of model is used to generate pictures from text, it can model pixel-level and generate paintings with significantly improved resolution in a shorter time. , The fineness of details has been significantly improved, and the finished products of some AI paintings have been comparable to the quality of human paintings.

Li Yanhong, founder of Baidu, concluded: “Artificial intelligence is moving from understanding content to generating content.” He believes that in the past year, artificial intelligence has undergone some directional changes. Baidu has been developing AI technology for many years. In the second half of this year, Baidu promoted its digital person Du Xiaoxiao to write essays for the college entrance examination and participate in art exhibitions. The works created by these AIs are all based on pre-trained large model technology, which can produce digital content at a tenth of the cost and a thousand times faster.

As someone who has been following the content industry track for a long time, Zhuang Minghao has a very obvious perception of AIGC’s learning speed. When the diffusion model was first discussed in April this year, the painting style drawn by AI was still a bit strange and hazy, but in October, it The resolution and fineness of the picture are improved, and the texture is so strong that it is difficult for ordinary people to see that this is a painting created by AI. In September this year, the AI ​​painting “Space Opera” won the Colorado State Fair’s digital art creation competition, and the judges said afterwards: “Although we didn’t know it was an AI work at the time of the competition, even if we knew, this award would still be awarded. “

When the paintings created by AI and the top human painters can be successful, AI video has become a new breakthrough direction. In October of this year, Meta and Google successively released their latest products for AI video. Among them, Phenaki of Google can use a series of text instructions composed of hundreds of words with logical relationship before and after to generate a feature film of more than 2 minutes, full of camera sense, rich plot and transition. Netizens who saw it exclaimed on social media, “Short video creators are also in danger!” “This thing is developing too fast, just like Pandora’s Box, once you open it, you can’t stop.” Zhuang Minghao said.

He feels that the current first stage of AI painting war has come to an end with the open source of Stablediffusion. “Its model is fast and good, high quality and open source, and anyone can use it.” Where will the technology break through next, and where will the industry be applied, AI video? Is the picture infinite expansion? Or 3D modeling? Expected yet unknown.

Commercialization Pathfinder

Painter Xiao Tao also participated in the excitement of AI painting. As a professional, she knows that AI painting is earlier than the public. In August, she tried several AI painting platforms at home and abroad. She told reporters that AI painting can improve efficiency and draw grand scenes beyond human imagination. She does not exclude this matter, but if the artist wants to rely on AI for commercialization, it is not possible at present.

“It is difficult to directly use the things drawn by AI in the works. The details are not good, the perspective is wrong, and it is difficult to make the exact same thing as you imagined.” Moreover, with the increase in the accuracy of the picture, AI needs to draw when painting. The keywords of the description are getting longer and longer, sometimes like writing a paper, but it is more troublesome. Xiao Tao feels that AI painting can provide some reference for the artist, but it is impossible to use it directly.

More people who have no foundation in painting are also joining in the excitement of AI painting. A person in the Internet industry told reporters that after seeing many friends sharing his AI paintings, he couldn’t help but try it out. He felt that he had discovered a new world and had many paintings created by himself. However, when the reporter asked him if he would pay for AI painting, he immediately denied it, “It’s just for fun, and I won’t take it seriously.”

At the beginning of YUAN, the people who were willing to pay for AI were mainly those who could make money from it. In addition to the novelists mentioned by Zhang Shiying, designers, painters, advertising and marketing companies and e-commerce merchants on YUAN Chu and imgcreator are the main paying people.

Among them, the enthusiasm for AI creation in the marketing industry exceeded her expectations. With the current ability of AI to paint, it can quickly generate pictures based on text, and there is still a gap between the overall texture and the works carefully created by humans. But for the marketing industry, they need a lot of pictures with cheap materials, and instead apply the pictures made by AI. “Many people are not so demanding. They are more willing to lay out 100 materials and wait for a certain material to explode before refining it. In the past, the most difficult thing was to find this explosive point. Now they can do ABtest at a very cheap cost. “Zhang Shiying said.

Although popular, the current use of AI for painting is still a small group. When the overseas market was the hottest in August, the number of midjourney users was in the millions, which was not large. According to the “China AI Painting Industry Survey Report” released by AI painting website 6pen in October this year, 60% of users have not paid for the use of AI painting products, that is, they are completely free, and the remaining 40% of paying users Among them, 16% paid less than 10 yuan, 14% paid less than 100 yuan, and less than 10% paid more than 100 yuan.

“The entire AIGC industry is still very early today, but there is a phenomenon worthy of attention that the industry is developing too fast.” Zhuang Minghao told reporters that if calculated according to the original cycle theory, it may take 3 or 5 years to achieve These are the scales now, but in the AIGC industry, these changes can be achieved in a year or two. After the stable model was open-sourced, some companies in the United States made plug-ins based on it, and the next-generation photoshop already supports AI painting. When an industry has companies that do the underlying technology and companies that do the application layer, it will become richer, the industry chain will become longer, the division of labor will be clearer, and the business model will be clearer. AIGC is moving in this direction.

In June of this year, Zhuang Minghao discussed the topic of AIGC with people from Baidu Wenxin’s big model. At that time, Wenxin people were still anxious because they didn’t know what was the use of this product, how to commercialize it, or how to commercialize it. Whether to do the B-side or the C-side, I don’t know if I should find a partner to do it together. By October, they had no anxiety at all, and countless people were pushing them forward. “They’re probably the least anxious now. It’s only been three months and so much has changed.”

After overseas AI painting reached its peak in August, the popularity of the user side is declining. The country is hot right now, but those who join in the fun will eventually leave. By then, will AI painting dissipate like a gust of wind, and will it really have commercial value?

Zhang Shiying has also been thinking about this question, “When will users really need and stick to AIGC? It must be integrated into their daily work. Therefore, it must be industrialized.” She told reporters that if it is just AI for painting, it may be more like a Interesting toy, “ToC is just a process of educating users, let everyone know that this thing exists in this world.” In terms of business path, she is more optimistic that AIGC can be used in design drawing, illustration, game, e-commerce and large scale. The potential of the hot metaverse, digital and virtual reality technologies.

Controversy and doubts

In the process of AI painting becoming popular, voices of doubt gradually emerged. From the end of August to the beginning of September, a large number of Japanese painters resisted AI painting on social media. They felt that the paintings created by AI imitating their style would damage the artist’s own image, and they said in their personal statements, please don’t do this in this kind of painting. Use my work in AI-generated services.

There was also a big debate on Weibo in October. Well-known illustrator Lao Lei said that he did not oppose AI painting, believing that it would only replace “low-end” painters, causing dissatisfaction among many painters. Later, painters such as MMMMM_Heye, Hua Chuicai, Wuhe Qilin, etc. also joined the debate, further breaking the circle of AI painting.

Xiao Tao understands the resistance of Japanese painters very well. The source of the incident was that an AI painting website in Japan launched a beta version, allowing users to upload 15 to 100 images for AI to learn, and then output AI paintings of the same style. “It’s not easy for an artist to form a mature style of painting. It’s really unreasonable to use other people’s things without paying a little cost.” She felt that the website’s approach was a bit excessive.

But she also supports the development of AI painting. “The advancement of technology is definitely a good thing, and it can improve the efficiency of painting.” She felt that in the early stage of technology, a hundred flowers must bloom first, and then the waves should wash away the sand, and it should not be banned in the early stage. But how to solve the copyright problem of AI painting requires attention.

Zhang Lijin also discovered a lot of AI copyright issues. She drew a picture on a domestic AI painting platform with the keywords of “The Lion King, City in the Sky”. The images of The Lion King and the animation “The Lion King” appeared almost Exactly the same, but with smaller eyes. “If the lion king’s eyes are replaced by a small one, is it not the lion king?” If these problems are not solved, it will limit the development of AI painting.

At present, there is no uniform regulation on the copyright ownership of AI paintings. Foreign midjourney platforms show that works generated by paying users can be used for commercial use, while non-paying users can only copy them for non-commercial purposes. The relevant person in charge of Baidu Wenxin Yige told reporters that Wenxin platform supports original paintings to enjoy copyright according to law, and encourages AIGC platform to continue to invest in intellectual resources, improve the quality of content, and protect the fair and legitimate rights and interests of all parties.

According to copyright law, the author of a work is a human being and nothing else. Zhao Hu, a partner of Beijing Zhongwen Law Firm and chief lawyer of the Intellectual Property Department, told reporters that there was a well-known case before. A monkey used a camera to press the shutter to take a good photo, but because the author is not a human being , so the work is not protected by copyright.

But the problem with AI works is more complicated. In the process of AI painting, someone made a language description, and the product of AI painting was also made by humans. Should the person who painted or the painting platform have copyright? Zhao Hu told reporters that the current mainstream view is that AI works do not enjoy copyright and are not protected. But there are also different perspectives emerging. “At present, various views are still colliding, and no conclusion has been reached, there is no clear provision in the law, and there has not been a particularly typical case yet.”

In October of this year, AI painting was hotly debated, and another reason was that some painters were worried that their work would be replaced by AI. In August of this year, the American “Atlantic Monthly” did not use the paid illustrations of the gallery website, but used a free image generated by AI to accompany the article. This illustration also aroused the dissatisfaction of the painter.

“When AI was first born, people said that simple and repetitive labor would be replaced by AI. Now, even creative labor such as artists is no longer an iron job for human beings.” Zhang Lijin was very emotional.

The concerns of current painters about AI are a bit like the concerns of the generation of painters when the camera was born in the 19th century. At that time, painters pursued lifelike portraits, but after the advent of cameras, everyone could take lifelike photos. The relevant person in charge of Baidu Wenxin Yige believes that with similar logic, AI painting will also bring about two changes: First, as a new tool, it will make the art creation process AI co-evolution, such as AI generating the first draft A new creative way of post-artificial modification; second, AI painting itself may form a new artistic genre.