AI is "occupying" the field of content creation.
Author | Zhou Xiaoli
Editor | Lizi
Some time ago, the work " Kill that Shijiazhuang People " by a well-known domestic rock band Universal Youth Hostel became popular on Bilibili. The reason is very special: every lyric of this song is matched with an AI-generating art tool called "Midjourney".
AI image generated based on the meaning of the lyrics
Since then, "AI painting" has spread like a virus on B station. Soon, songs such as " Qilixiang ", "The Lonely Brave", "California Hotel", and "Bohemian Rhapsody" have all become excellent materials for AI painting without exception.
is more famous in August this year, an art work called "Space Opera" won the first prize at the Colorado State Fair in the United States. Later, its author Jason Allen revealed that the work was made by AI.
award-winning work "Space Opera House" created by Midjourney through AI drawing tool Midjourney
Interestingly, the competition judges did not rejudgment on the work and bluntly stated: Even if it is an AI-generated work, it is still worthy of such results. But obviously other artists are indignant about this, believing that this is a "plagiarism" of creativity by high-tech.
Once upon a time, people scoffed at the prospect of AI in artistic scenes, believing that AI can only complete some calculation work that relies on algorithms, and artistic creation is a unique talent given to mankind by God. And now it is obvious that this last pride of human beings is also being "swallowed" by AI.
In the past two years, various AI painting platforms such as DALL·E 2, GauGAN2, Stable Diffusion, Midjourney, etc. have emerged rapidly. Last January, OpenAI launched DALL·E. Just one year later, the upgraded version of its upgraded version of DALL·E 2 started to generate more realistic and accurate images at 4 times the resolution, and the number of registered users exceeded 1 million in less than 3 months.
What is popular is not only AI painting, but also AI-generated videos. Not long after the official open source of Stable Diffusion, Meta has brought the new product Make-A-Video, which can generate videos directly through text. Subsequently, in less than half a month, Google entered with AI models from text to high-definition videos Imagen Video and Phenaki (the former tends to create video quality, while the latter tends to have video logic and duration) .
It is obvious that with the breakthrough of artificial intelligence generation capabilities, content production has entered the era of artificial intelligence generated content (UGC) from user-generated content (PGC) . The emergence of AI painting to AI video indicates that the AIGC era has begun.
1. Why are technology giants all planning AIGC?
For AIGC, founder, chairman and CEO of Baidu Li Yanhong explained at this year's Baidu World Conference: AIGC is "autonomously generated content from artificial intelligence."
In Li Yanhong's judgment, AIGC will usher in three stages of development:
The first stage is the "assistant stage", and AIGC assists humans in content production;
The second stage is the "collaborative stage", and AIGC appears in the form of virtual human coexistence of reality, forming a situation of human-computer symbiosis;
The third stage is the "original stage", and AIGC will independently complete content creation.
In fact, AIGC is not a new thing. It has been discussed for a long time, such as Microsoft Xiaobing and other artificial intelligence products that write poems, write, and create songs, but there has never been a large-scale popularization of standardized to C products.
But why, a few years later, AIGC began to explode again, attracting technology giants to make plans?
According to the qubit AIGC map, AIGC is now mainly used in text, images, video, audio, games and virtual people. Most of the startups involved are concentrated in rounds A to B, including DeepMusic, Reflection Audio, Lingxin Intelligence, Color Cloud Xiaomeng, rct AI, Film and Streaming Technology, Hyperparameters, etc.
China's AIGC industry chain division Source: "AIGC/AI content generation industry outlook report"
, and domestic major manufacturers Baidu, Tencent Youtu , Alibaba , Kuaishou, ByteDance, NetEase, SenseTime, Meitu, etc. have also made some investments in the AIGC field.
For example, Baidu launches AI art and creative auxiliary painting platform Wen Xinyige; Tencent builds writing robot "Dream Writer"; Alibaba's online design platform Lubanner helps marketers produce Banner; ByteDance's Jianying and Kuaishou Cloud Clip can provide AI-generated videos; NetEase launches a one-stop AI music creation platform "NetEase Tianyin", etc.
Foreign countries are fighting against gods in the field of AIGC. There are technology giants Google, Meta, , Microsoft, , etc., as well as AIGC's new unicorns Stability AI, Jasper, OpenAI, etc. And technology companies soon continued the popularity of AI painting to AI-generated videos. From Meta's announcement of the system Make-A-Video from text to video, to Google's announcement of Imagen Video and Phenaki that can generate high-definition videos from simple text prompts, AIGC is growing rapidly overseas.
An important reason why major domestic and foreign companies have entered the AIGC field is the official open source of the text-image generation model Stable Diffusion.
This time, Stable Diffusion is open not only to the programs, but also to its trained models, which means that successor entrepreneurs can better use this open source tool to explore a richer content ecosystem. The open source of Stable Diffusion plays a crucial role in popularizing a wider range of C-end users.
Secondly, the popularity of AIGC is also due to the rapid development of technologies such as generating diffusion models and multimodal pre-training models, and has made significant progress in the generation effect of graphics and texts, allowing AI to quickly and flexibly generate data content of different modes.
Before 2021, AIGC will mainly generate text. The new generation of models can handle any content format, including text, voice, code, images, video, 3D models, robot actions, etc. For example, the recent AIGC technology represented by DALL-E2 and stable-diffusion can be widely used in content generation, editing and creation in terms of graphics and text generation effects. Wan Pengfei, head of the Y-tech AI Technology Center of Kuaishou, told "Jiazi Light Years" that a major advantage of generative technology is that it can not only improve the efficiency of content acquisition and editing at the tool level, but also provide reference for people at the creative and strategic levels. (Note: Generative technology: that is, the technology that uses existing text, audio files or images to create new content. By generating AI, computers detect basic patterns related to input and generate similar content)
At the same time, the popularity of various social and streaming media platforms today is also driving the evolution of content production methods. AIGC is a new generation of content production method after PGC and UGC. The underlying reason is definitely demand-driven. As people's demand for content becomes more and more powerful, the content industry must also upgrade and iterate. This has gradually evolved from the tools that assist content creation in the past to being able to create directly, and has been able to do many creative categories such as writing, painting, composition, and design.
Finally, there are external environmental factors. In the economic downturn, the technology industry will choose corresponding convergence expenditures and focus on more pragmatic places such as the commercialization of artificial intelligence.
Under the epidemic, enterprises have emphasized cost reduction and efficiency improvement, and artificial intelligence generation technology has therefore become the first tool for creators and teams to enhance their creative capabilities. So when the market is sluggish, this business will accelerate its explosion.Just like every financial crisis, some entertainment-related Internet companies will appear.
international consulting company Analysis Group released a report saying that by 2031, the metaverse's contribution to global GDP can reach US$3 trillion. The digital life ecology built around the virtual world and meta-universe infrastructure and the music ecology built around the new Internet media have taken shape. It is only a matter of time before AI technology can be implemented on a large scale.
2.AIGC is booming, a new round of industrial evolution is coming
After an industry is booming, the first thing that smells the "fragrance" will always be capital.
On October 19, Jasper.ai, an AIGC company that focuses on text generation, announced that it had completed a US$125 million Series A financing, with a valuation of US$1.5 billion. It only took 18 months since Jasper AI was launched.
. Just the day before Jasper.ai announced its financing, Stability AI, another leading company in the AIGC field, announced that it had received US$101 million from Coatue and Lightspeed. The company announced that it would continue to develop AI generation models for generating images, languages, audio, video and 3D. The post-investment valuation reached US$1 billion, becoming the new unicorn company .
However, compared with the situation where several unicorns have appeared in overseas markets, the Chinese venture capital circle has not really become "hot" in taking action. As of now, AI painting startups that have entered the financing stage can be publicly found in China, and only TIAMAT was awarded a multi-million-dollar angel round investment in DCM in October. Other companies or platforms in this field, such as 6pen, draft.art, great painter Domo, Dream Inception, etc., have not entered the financing stage. Gao Ning, a senior investor in
, has been paying attention to the AIGC field recently. When he was communicating with "Jiazi Light Years", he said that AIGC is indeed a focus of the capital market in recent times.
He believes that if you start a business in the AIGC field, it is best to start a global market, because there are inevitably Chinese or Chinese cultural carriers, and many opportunities will be born.
China's content industry is huge in scale and has many fields, including a Chinese online literature market with more than 500 million users, a Chinese comic industry with a market size of over one trillion yuan, a Chinese advertising industry with a market size of over one trillion yuan, and a Chinese media industry with a market size of over 3 trillion yuan.
In the context of the current recurring COVID-19 pandemic, the demand for digital content is also stronger. Sequoia Capital expects generative artificial intelligence to "generate trillions of dollars in economic value."
In fact, with the accelerated improvement of global informatization level in recent years, the integration and development of artificial intelligence and media industries have been continuously upgraded. As the current new type of content production method, AIGC has taken the lead in achieving major innovative development in industries such as media, e-commerce, film and television, and entertainment, which have high digitalization and rich content needs.
In addition, under the promotion of the integration of digital and actual numbers and accelerate industrial upgrading education, AIGC applications in various industries such as finance, medical care, and industry are also developing rapidly.
Artificial Intelligence Generation Content (AIGC) Application View Source: China Institute of Information and Communications
Overall, under the current trend of the merger of digital economy and the real economy, the combination of virtual self and real self, the prerequisites for the development of AIGC are already in place, which has greatly driven the development of related industries.
- Entertainment & Film and Television Industry: AI helps video script creation, create virtual idol IP, etc.
Since September this year, Meta and Google have successively announced their latest achievements in the cutting-edge field of AIGC. In particular, Phenaki, an AI video generation model launched by the Google team, can generate variable-length videos based on text content. In the announced DEMO, Phenaki composed a logically coherent video based on hundreds of words in just two minutes. It can be seen that Phenaki is aiming at long video production. The emergence of Phenaki will inevitably have an impact on the entire video industry in the future.
At the same time, the use of AIGC technology can effectively stimulate the creative inspiration of film and television scripts. AI virtual digital people can also appear in film and television scripts to play different roles, greatly improving the post-production quality of film and television products in short dramas, and helping film and television works to maximize cultural and economic value.
- E-commerce industry: digital people assist in selling goods, XR product display, etc.
Currently, AIGC is widely used in the e-commerce industry. By creating virtual anchors, e-commerce can provide audiences with 24-hour uninterrupted product recommendation introductions and online services, and the threshold for merchant live broadcast is therefore lowered.
In addition to using digital people to assist in e-commerce, digital people is also used in scenes such as film and television creation, animation, VR\AR\MR, TV hosting, virtual idols and other scenes.
"Jiazi Light Year" learned that in the field of AIGC, many companies choose to put their implementation scenarios on digital people, including major Chinese and foreign Internet companies such as Amazon , Google, Apple , Microsoft Xiaobing, Baidu, Tencent, and many startups.
Digital people are a track that has only appeared in the past two years. The competition is far from being as "volume" as TTS (voice synthesis technology) . Currently, most domestic digital people are still in the early stage, which also means to a certain extent that the opportunities of startups may be hidden in a more vertical application track, and finding the right direction is very important.
Reflection Audio is a company that provides virtual digital human technology solutions. Through neural rendering technology, it has created the AI digital clone of ophthalmologist Tao Yong , realizing the implementation of AIGC in health science popularization scenarios.
can generate popular science audio/video content production methods by inputting text. Neural rendering technology has fully liberated the real labor force of medical experts. Compared with traditional 3D modeling methods, neural rendering technology can create AI digital clones more quickly, reducing time and financial costs, which allows AI digital people to have a wider implementation scenario and is easier to sink to C-end users.
According to Wan Pengfei, in the next 1-2 years, digital human + AIGC will be a more promising commercialization direction. Digital humans are a new human-computer interaction and everyone-to-people interaction mode, and AIGC is a new content production mode. The two concepts can be combined and unlocked many valuable application scenarios. They can be widely used in entertainment live broadcasts, e-commerce live broadcasts, video production, digital employees, virtual idols and other fields.
- Advertising & Media Industry: Creativity and material generation, virtual world interaction, etc.
With the addition of AIGC, the creator economy of all walks of life has ushered in new growth points. For example, the dubbing industry of audio books, the film and television dubbing industry, the animation artist, or the designer of a marketing advertising company may be the main users of AIGC in the future to assist their industry in performance optimization.
In addition, many media organizations have begun to use AIGC-generated pictures as magazine covers, and some writers or novelists can also use AI to draw pictures for their articles or novels. "The Economist" used pictures generated by Midjourney as the magazine cover some time ago. The AI-generated pictures will be further popularized in various industries.
Magazine cover made by economists using pictures generated by Midjourney
- Medical industry: AI intelligent diagnosis and treatment, human-computer emotional interaction
In the field of AIGC, there are currently not many companies in the vertical track. In addition to the relatively mature financial retail and customer service tracks, mental health is one of the most promising industries that are deeply integrated with AIGC. However, although the mental health track is large, due to the high ceiling, most AIGC companies are unable to leverage their technological advantages due to the limitations of the integration of professional fields, and the supply of high-quality solutions in the industry is seriously insufficient.
In fact, through AIGC's technology and means, the medical industry can achieve standardized and effective intervention and treatment on the supply side.For example, virtual people can be used to imitate psychotherapists or doctor assistants, use AI-generated dialogue to establish a foundation of deep trust with users, and then achieve therapeutic effects through role-based and personalized communication.
Lingxin Intelligence, founded by Huang Minlie , a computer professor at Tsinghua University, is a typical AIGC company. It has been deeply involved in the mental health industry for many years and has accumulated a large amount of Chinese dialogue data. Based on the big model, it has built a unique model framework at multiple levels such as emotional support, listening and companionship, role-playing, and open chat. It has entered the mental health track through industrial application logic driven by generative dialogue models and other AIGC capabilities as the expression form.
The dialogue robot "Emohaa" developed by it is mainly used to build an interactive digital diagnosis and treatment plan with AI-generated dialogue as the core, allowing the robot to express its understanding of users and empathize with , and provide timely emotional support and psychological counseling to achieve good treatment and recovery results.
- Game industry: Game NPC character generation, scene and level generation
Game industry can use text generation capabilities to create rich and interesting game NPC capabilities. According to different scenes in the game, the corresponding NPCs are set, and all NPCs' answers can be generated in real time according to the prompts for setting words. In addition, using AIGC for the creation of some micro-materials is also a short-term feasible opportunity.
However, from the current development of AIGC in the industry, the biggest problem is that the industry has not yet established a clear monetization method.
Taking writing robots, automatic dubbing, AI painting and other scenarios as examples, most products are still in the free trial stage of traffic attracting, and the charging space is small; and most of them are lightweight tool products, which do not have larger content scenarios. Whether it can effectively contact C-end users while the Internet traffic is relatively stable and achieve good activity and retention rates is still a challenge.
But this also brings broad space for growth for AIGC, allowing it to move towards a larger industrial direction of social or content community.
3. Let demand drive, rather than technology force
At present, whether it is a giant Internet company or a startup company, they are gradually exploring the direction of AIGC. Currently, it is mostly concentrated on the AIGC direction at the perception level to explore the commercial implementation scenarios above.
Huang Minlie told "Jiazi Light Year" that from an industry perspective, although foreign countries are relatively ahead, some more typical companies mainly do visual perception intelligence levels, such as the generation of text to pictures or text to videos.
Huang Minlie believes that the commercialization of the AIGC field in may develop in three levels in the future.
The first level lies in the perception level , that is, something that is directly and simple and can bring sensory stimulation. In the early stage, we focused more on perceptual intelligence at the audio and visual level, including AI mapping and composition, AI video, 3D, etc.; the second level of will come to the cognitive level , which will gradually become a trend related to dialogue, writing, error correction, and language generation; the third level of is to develop a new ecological chain for specific industries, providing high-quality and complete solutions for the entire industry.
Based on the current situation, Huang Minlie said that in the future, AIGC can consider relating to specific scenarios and specific application directions, that is, developing towards a combination of scenarios and industries, which can make AIGC a good auxiliary tool and empowerment method.
Judging from the hottest AI painting now, the threshold for painting creation is constantly lowering. Just click to enter the server that generates the AI painting official website, and then enter or call the "/imagine" option in the chat box and enter the scene text you want to describe in your mind.
AI painting emphasizes the generation of new content rather than auxiliary analysis and decision-making of historical data. Painters can use it to help painting, and use it to draw character designs, and independent game producers can also significantly reduce costs through AI painting.
on the Wenxin Yige platform, which inputs "no wind, rain, no sunshine" to create pixel-style painting
Gao Ning believes that pictures themselves are a killer application. Although the commercialization of sound or text types may not be done poorly, in terms of the propagation effect, the visual impact brought by images is also one of the reasons why this application is truly popularized.
Similarly, Chenshan Capital Wu Wenchao also believes that after the Internet era begins, marketing is one of the most important monetization methods for traffic enterprises, and an important way to carry marketing is rich media content such as graphics, texts, videos, etc. with more visual impact.
From AI to AI-generating images to AI-generating videos, the computing power requirements are exponentially improved, which allows chip manufacturers who can provide GPU computing power to taste the sweetness. The advanced graphics processor produced can become an ideal choice for training and deploying artificial intelligence models.
Just a while ago, Nvidia CEO Huang Renxun publicly stated that generative artificial intelligence is a key use of the company's latest chips, and these programs may soon "change communications."
At the same time, the large amount of content generated by artificial intelligence generation companies will also promote the development of the cloud computing industry.
cloud vendors hope that enterprises can apply on the platforms and frameworks they build. It is reported that Meta and Google have hired many professionals in the field, hoping to integrate this advanced technology into the company's products; while Microsoft will add DALL-E to its Office suite and Azure AI, and Adobe plans to add the generated AI tools to Photoshop. For small and medium-sized companies, while using cloud services, they can also use the AI systems provided by these platforms to complete their required functional architectures.
With the development of artificial intelligence technology, capital players are accelerating the implementation of various AI applications, and virtual digital people, automation applications and other products are emerging one after another. In the field of "AIGC+art", AI generation of pictures, text, audio, video and other content has gradually penetrated into literature, painting, short video, education and other scenarios, further broadening its commercialization space.
However, from the product perspective, many content generation tools that appeared in the early stages of AIGC are more of C-end products due to the low threshold for use, but most players are only for entertainment and it is difficult to convert into real paid users. If AIGC is used on the B-side, it can assist workflows or actually improve work efficiency, whether it is used to write marketing copy or generate image, it can generate value, and it can become a sustainable business model.
Another difficulty is that even if various AI painting companies have formulated calculation standards for to B or to C, it will be difficult to make money by painting work in the short term, because the training cost of the model is too high. Many domestic painting platforms use self-developed models, which increases the training cost for AI painting tools. For many start-ups, this is a considerable cost.
. For enterprises, the essence is always profitable. Only when phased artificial intelligence achievements have large-scale commercial applications can they bring value to enterprises, otherwise they can only stay in the encirclement and self-entertainment model.
Wu Wenchao said that no matter domestic and foreign unicorns or startups, AIGC does not have a mature business model in commercial monetization, which is very different from the original technology driven by the industrial field.
"For example, CV has a particularly clear scenario that requires facial recognition. Even if AI does not appear, it still has facial recognition scenarios. However, with AI, facial recognition can make facial recognition more accurate."
But from the perspective of content generation, it is essentially a creative industry. There was no such demand in the industrial field before. It is generally believed that both design and 3D models require manual labor, rather than endogenous demand in the industry.
Wu Wenchao described that now, using technology to find demand, is a bit like using a hammer to find nails.This means that the development of AIGC is driven by artificial promotion compared to the original industrial demand in the CV field, so the business model will not be clear enough.
Although each company has different model architectures, it is technically similar. How to productize the next step, how to build a good community, how to better implement user feedback on the model, and implement some to B segmented scenarios at the same time, is the real difference between them and competitors.
4. In the stage of disorderly development, the bullets must fly for a while.
In recent years, with the support of big data and large computing power, the research on artificial intelligence has also been pushed to a new level. Jasper CEO Rogenmoser believes that "every tool in the world will build up artificial intelligence with some ability."
From the perspective of technological development and evolution, every technological change will interweave ethical challenges. At present, the laws, regulations and ethical norms of artificial intelligence have not been formed, and ethical issues will become a great resistance to the development of artificial intelligence.
Image source: Getty Images
At present, the ethical issues about AIGC are mainly reflected in the inability to confirm rights and copyright disputes. Although the development and revolution of technology can bring about a certain extent the prosperity of content, market returns come from market transactions, and the basis of market transactions lies in rights confirmation. If AIGC cannot confirm the rights, on the one hand, it will lead to the infringement not being effectively resolved, and in addition, it will also reduce capital's enthusiasm for investment in the AIGC field accordingly.
In the copyright dispute, some plagiarists have been seen appearing, trying to use Stable Diffusion's open source technology to do the same productization. In addition, it is also difficult to define whether the pictures generated by AI paintings are imitating the artist's style.
With the development and progress of AI technology, the level of automation will continue to be improved in the future and will be more closely integrated with reality. The explosive growth in the AIGC field will aggravate the issue of data privacy and ownership.
Judging from the most discussed AI painting and AI generation videos, as the generation tools gradually move towards the public and commercial markets, the risk of content fraud is getting higher and higher. On many NFT platforms abroad, direct AI-generated works have been sold endlessly. In China, if you search for AI painting on platforms such as Taobao and Xianyu, you will find that many people are using AI painting to make profits. However, since AI works at have not yet been clearly defined in the copyright and legal level, it still wanders in the gray area.
For this reason, the relevant person in charge of Wen Xinyige told "Jiazi Light Year" that since AI can generate images of comparable quality to human paintings, it is necessary to support original paintings to enjoy copyright in accordance with the law. It is recommended to formulate implementation rules in classification and hierarchy based on the governance of innovative business applications including AI painting based on different subdivided application scenarios and product models.
However, since the industry is still very new and the policies of different countries are different, each AIGC company has different methods of handling AI creation. Therefore, the whole world of is still in a game process.
For example, in terms of security, Open AI directly filters out many security words, prohibiting the appearance of some specific characters and political relatedness; Getty Images prohibits users from uploading generative AI images to their inventory image database; TIAMAT and Midjourney unanimously chose to use copyright-free images in the material library to avoid copyright disputes.
Regarding the ethical problems faced by AIGC, Huang Minlie said that this is an inevitable problem in the technological development path. This is because the current big model AI capabilities are easy to remember and imitate, but they cannot be created, so they will inevitably face some copyright and even anti-human ethical problems.
However, from the perspective of the entire technological development, we should still be optimistic. He said that we can let the technology develop for a period of time, let the technology expose the problems, and then find ways to constrain and regulate it from the perspective of policies, laws and regulations to form a better development.The overall goal is to adhere to the premise of allowing technology and AI to serve humans, ethical issues can be solved in the future.
OpenAI current CEO Sam Altman also said on Twitter, "AI will bring huge changes to the world, and we should change the economic system to adapt to it." AI is still developing at an accelerated pace, and there will be more boundaries in the future, and regulations will continue to be improved. AI development and regulations will always be adjusted to .
Just like the two sides of a coin, it does not deny that the development of AI technology has made human beings more productive and efficient. Then, let the bullets fly for a while.
