Shulou(Shulou.com)11/24 Report--

With a word, AI can generate a creative painting for anyone interested in creating art, regardless of whether the creator has received professional artistic training.

In the past two years, overseas users have gradually become a trend to create art with the help of AI painting software such as Disco Diffusion and MidJourney. In China, there are also AI art and creative auxiliary platforms suitable for Chinese descriptor input and better understanding of Chinese cultural aesthetics, such as Baidu Wenxin Yige and TIAMAT.

For professionals, these AI drawing aids are a good creative assistant. Some amateurs have used these tools to develop their own slant careers, making wallpaper numbers on chatter, selling descriptions on idle fish, and selling their works to packaging factories or digital collectors to earn some pocket money.

01. AI-style creative freedom Mr. Wu is a doctor in his job. He likes to play chatter. In his spare time, he often thinks about doing something on chatter to make money.

Two months ago, he saw a lot of creators doing wallpaper numbers on the chatter, that is, creators would post some good-looking photos or pictures that users could download and use as desktop computers or mobile phones. When downloading, users need to see a platform advertisement, and creators can get platform shares from advertising.

According to Mr. Wu's observation, the wallpaper number circle is still a minority, but there are also 200,000 or 300,000 fans in the circle. By delivering beautiful photos such as scenery, characters and cars to users, tens of thousands of people enter the monthly income.

Mr. Wu himself is a landscape photo enthusiast, also want to do a landscape wallpaper number, but suffer from not much material. When he was worried, he came into contact with AI painting and found that he only needed to input relevant descriptors in AI painting software to generate good-looking pictures, which solved his problem of lack of materials.

Mr. Wu downloaded two overseas AI painting products, one is Disco Diffusion and the other is MidJourney. He registered an account named "AI Beauty Picture" on the chatter and started painting.

DiscoDiffusion is free for three hours a day, and MidJourney Wu charges a membership fee of $30 a month. The creative efficiency of each picture can be generated in an average of a few minutes, which is enough to support Mr. Wu's material needs for his account.

In Mr. Wu's view, the two software are good at different focuses. Disco Diffusion is good at drawing scenes, MidJourney is good at drawing close-ups of people or animals.

Fig.| Mr. Wu created works with descriptions such as Dream Moon, Milky Way, hanging garden under night sky, magic and light, which have been authorized by the creator. With the blessing of these two software, Mr. Wu, as a doctor, realized his creative freedom.

He was able to publish seven or eight works a day by typing in various descriptors to try and make mistakes. Although Mr. Wu later discovered that because the user circle is too small, there is no need to upload so much in the chatter every day.

What is more valuable is that although there is no artistic accumulation, it does not prevent Mr. Wu from creating works that can be sold at a price.

According to reports, Mr. Wu created a cyberpunk robot, the description behind it, was bought by a person engaged in digital collection trading for dozens of yuan, which surprised Mr. Wu.

Later, this person asked Mr. Wu to create a group of zodiac pictures for digital collection development. Mr. Wu used AI to create three pictures for each group of zodiac animals, and the description behind the last group of pictures was sold for hundreds of yuan.

This inspired Mr. Wu. Because Mr. Wu's chatter number fans are only more than 200, it is difficult to earn money through advertising sharing, he simply decided to sell descriptors.

A hundred descriptive words are priced at 20 yuan, get chatter, idle fish to sell, there are printers to buy, this formed Mr. Wu's pocket money income.

Similar to Mr. Wu's experience, there is also a small red book blogger, earthy cowboy, a photography professional, who began trying to use MidJourney to paint AI since June this year.

The AI painting works of earthy cowboy mostly take cowboy as the theme, the style covers oriental elements, future feeling, science fiction feeling, etc., and the number of praise is thousands. He is used to telling AI the brain hole that cannot be realized when photographing, so that AI can help him realize it.

Fig.| Cowboy created by earthy cowboy, released with authorization from the creator. In four months, the earthy cowboy released 117 AI paintings in the Little Red Book, among which the average number of times behind each work was 50, which can be described as excellence.

Creators are learning, AI is learning. At present, MidJourney has recorded more than 10000 generation records of earthy cowboy. Now, as soon as you type in a relevant word, the AI can generate additional associations. Sometimes earthy cowboys miss AI that "doesn't listen to people" and is more creative than sleek AI that is getting closer to human thinking preferences.

After publishing AI paintings on Little Red Book, Earthy Cowboy said he had received several invitations, and some people came to the door to buy the original pictures directly for NFT collection, but there were only a few such people. His current income from AI paintings is not enough to cover his investment in MidJourney.

According to the observation of earthy cowboys in the circle, the reasons for using AI painting vary. Some illustrator friends will use AI to generate drafts of their works, which will be used as a basis for creativity and modification. Friends who do architectural design will use AI to brainstorm. But more plain people are trying to be fresh, just want to see "what Superman and Iron Man look like together."

Driven by users, there are more and more videos introducing AI painting software and tutorials at Station B and Shaking Sound. Little Red Book has become an important display platform for AI painting works, and idle fish plays a platform role of selling descriptors.

The origins of AI painting can be traced back to the 1970s.

Ten years after the advent of computers, artists created computer programs AARON to create paintings. The main principle was to control robotic arms to paint, but the technical cost was relatively high at that time, and only abstract results could be obtained.

Fig.| AARON and its inventor Harold Cohen and AI painting in modern sense mainly refer to AI learning the association between "language description" and "art picture" from massive data corresponding to graphics and texts, starting from randomized starting point, constantly correcting paintings, and finally forming finished products that are highly consistent with human experience and knowledge aesthetically.

According to Baidu Wenxin a grid related responsible person introduced, modern meaning AI painting technology rudiment, mainly appeared in the last decade.

At that time, the technology could complete the generation of pictures specified by text information in batches, including style transfer, element splicing and GAN-Generative Ad Serial Network.

Judging from the quality of the generated pictures, although these technologies have made significant progress in the level of AI painting and have preliminary application possibilities in relatively strong styles, there is still a gap with the quality of human painting.

The most recent technological explosion of AI painting was the introduction of diffusion generation models in 2022. After these models were used in Wensheng maps, they could model pixels and generate paintings with significantly improved resolution in a shorter time, and there was a significant improvement in aesthetics and detail fineness. Some AI paintings were already comparable to the quality of human paintings.

It is generally believed in the industry that from 2021 to 2022, AI Wensheng technology has developed fastest in two years, many new learning models or products have been born, and the trend of AI painting has also begun to develop.

One is CLIP (Contrastive Language-Image Pre-Training), a deep learning model released by Open AI in early 2021.

Open AI, an artificial intelligence research company founded in December 2015 in the United States, has created two AI systems, DALL-E and DALL·E2, with CLIP, to build users the ability to create more realistic and accurate images based on text descriptions.

The second is Disco Diffusion, an open source painting program that began appearing and popular in early February this year. It is an AI launched by Google and developed by Somnai_dreams, an artist and developer.

Disco Diffusion is characterized by its ability to efficiently synthesize visual data, and the quality of image generation directly beats the GAN model. However, defects still exist, mainly manifested in the lack of detail in AI painting, the rendering time is too long, and the memory resource consumption is huge. Stable Diffusion, which began testing in July, quickly corrected these shortcomings.

The third is MidJourney, which was launched in June this year. It is a creative AI. The quality of the pictures generated by it has been greatly improved, and the commercialization in the form of monthly fee has been relatively successful. Public records show MidJourney has more than 2.4 million members.

Fig.| The paintings displayed on the MidJourney community are in China, and AI paintings are gradually gaining popularity.

Models such as Stable Diffusion and MidJourney are mainly aimed at users in the English world. Differences in language and culture form obstacles for domestic users to use overseas AI painting software. Domestic users call for Wensheng map products in Chinese world, which forms certain market demand.

TIAMAT was founded in 2021 and is known as the first AI generative platform in China. In the first half of this year, TIAMAT began publishing Chinese Wensheng pictures on social media such as Little Red Book and Shaking Sound, catching up with the first wave of AI painting.

Fig.| AI Painting on Little Red Book

Fig.| TIAMAT is dedicated to providing services in the form of creative production, advertising, digital collections, etc. for individual users and corporate customers. Recently, TIAMAT has announced the completion of a multi-million dollar angel round of financing from DCM.

In August this year, Baidu released AI art and creative auxiliary platform Wenxin Yige, which uses Baidu's independent research and development of industry-level knowledge enhancement large model-Wenxin large model, through semantic understanding + multi-painting style adaptation + result optimization sorting, so that users only need to input a sentence, can complete paintings, including intention, figurative and other types and styles.

Wenxin Yige is considered to effectively solve the difficulties of demand description, strong diversification demand and model generation result screening in the Chinese world, and reduce the difficulty of AI painting for users in the Chinese world.

Based on the platform of Wenxin Yige, the "Unbounded" series paintings created by Baidu Digital Rendu Xiaoxiao have been evaluated as "having met the basic requirements of undergraduate art students" by professors of professional art academies.

Fig.| Du Xiaoxiao created an unbounded series of insiders pointed out that for AI painting, how to make AI accurately understand the descriptive language in Chinese and English context is very important. Descriptive language not only has language differences in Chinese and English, but also cultural differences carried by language, thus affecting the research and development of AI painting technology.

For example,"dragon" basically has no ambiguity for English environment, but in Chinese environment, there are image differences between eastern dragon and western dragon, and the meaning of eastern dragon and western dragon as cultural symbols is completely different, which poses a great challenge to AI painting technology.

From the present point of view, how to deal with the differences of language and culture, AI painting still has a long way to go.

03. Controversial points and future development of AI painting After the birth of AI painting, a controversial point in the market mainly revolves around copyright.

The bottom logic of AI painting platform is to integrate the works of various creators and the works generated by existing Internet pictures, extract the elements such as brush strokes, colors and compositions of the original works, and carry out "secondary creation." It is difficult to define whether the copyright belongs to the original author, platform party or current creator at present.

According to users who have used Disco Diffusion and MidJourney, the copyright of images generated with Disco Diffusion belongs to the creator himself, while the copyright of images generated with MidJourney belongs to the creator and the platform.

Baidu Wenxin grid related responsible person thinks, AI technology generated pictures is not both picture patchwork, but similar to human after long-term painting training re-creation, both inheritance, but also have their own characteristics.

However, based on copyright issues, the industry is still under discussion. In order to avoid possible risks and solve possible disputes, Baidu Wenxin said that in the future, if the generated pictures opened by the platform infringe on the rights and interests of the original author, it will provide complaint feedback channels to provide rights relief channels for relevant obligee and deal with them quickly.

It is understood that there are no relevant specific legal cases in China that can be used for reference.

Fig.| Input the pictures generated by "artificial intelligence picture and surrealism" in Wenxin grid. The essence of the development direction of AI painting is the development of AIGC (Artificial Intelligence Independent Content Generation) technology. It is a brand-new content production mode after PGC and UGC, which is considered to stimulate the further development of content industry.

Public data show that the global stock picture market is close to $300 billion. But at present, the mainstream image market is what users use on the platform, not what users want. Content producers are still constrained by the imagination of production, and the traditional picture industry is still in the "digital handicraft" stage.

In contrast, AIGC evolved to accommodate content creators 'extension of their productive imagination. In the past, AI was only a tool to assist content creation, but now AI has been able to independently complete creative work such as writing, design, painting and composition.

One thing to note here is that AIGC sounds hip, but automated content generation is not a very new concept.

The use of computer-assisted human content production has been around for a long time, such as code hinting in IDE in computer programming, error correction in editing content with Office Word, and later automated text generation with NLG.

Baidu CEO Li Yanhong has a discussion on which stage of development AIGC technology is at at the 2022 Baidu World Conference.

He divided AIGC into three stages of development: the first stage is the assistant stage, where AIGC assists human content production; the second stage is the collaboration stage, where AIGC appears in the form of virtual human coexistence, forming a man-machine symbiosis situation; and the third stage is the original stage, where AIGC will independently complete content creation.

Judging from the current development of AI in various fields such as composition, painting and song writing, AIGC has entered the assistant stage, but it is generally in the early stage of development.

Whether AI painting will replace designers and original painters in the future is still uncertain. But judging by the reaction so far, people are clearly not that worried.

This article comes from Weixin Official Accounts: Zhengu Research Office (ID: zhengulab), author: Zheng Sifang, editor: Gong Zheng

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.