Recently, an AI-related case attracted the attention of peers across the country. The reason was that a Baijiahao author used an online AI picture as an accompanying picture when writing an article online. As a result, the producer of the picture was sued for infringement. This makes people nervous: can the pictures produced by AI be used casually in the future? Will it be infringement?
The thing is like this -
In February 2023, Li Yunkai used the Stable Diffusion model to generate several portrait pictures by inputting dozens of prompt words on the model, setting the relevant iteration steps, image height, prompt word guidance coefficients, and random number seeds. Later, Li Yunkai posted it on the social platform Xiaohongshu under the title "Spring Breeze Brings Tenderness". On March 2, 2023, an article titled "Love in March, in Peach Blossoms" published by a self-media account on Baijiahao used pictures produced by Li Yunkai.
Pictures generated by Li Yunkai using the Stable Diffusion model
Li Yunkai believes that the other party cut off his signature watermark on the Xiaohongshu platform for publication without his permission, which violated his signature rights and information network dissemination rights, and he should bear legal responsibility.
In May 2023, Li Yunkai sued the defendant Liu for infringement of the rights of authorship of works and disputes over the rights of information network dissemination in the Beijing Internet Court.
The most critical controversy here is whether the picture "Spring Breeze Brought Tenderness" produced by Li Yunkai using the Stable Diffusion model based on the prompt words and parameters set by himself constitutes a work? And what type of work does it constitute?
The court held that the pictures created by the plaintiff Li Yunkai using the open source software Stable Diffusion were completed independently and reflected his personalized expression. Therefore, the pictures involved in the case met the requirements of "originality". The pictures involved in the case meet the definition of works, are works of art, and are protected by copyright law. The final verdict was that Li Yunkai won the lawsuit, and the defendant was required to publicly apologize on the Baijiahao platform and compensate the plaintiff 500 yuan .
This case has also become the first copyright case in the field of AI-generated images, making practitioners and users in the AI field more vigilant:
It turns out that pictures created by AI also have copyrights. So in the future, when using AI pictures, do we have to pay attention to authorization issues?
Many AI pictures are "melted" out of other people's pictures, which is equivalent to "secondary creation." This approach has long been regarded as plagiarism in the field of literature and art. Why can it still enjoy copyright? Is this reasonable?
What are the rights and risks associated with the pictures I create using AI? How are rights and responsibilities divided?
In this regard, we invited Deng Chao, a professional lawyer in the field of intellectual property, to have a conversation and talk about what is going on.
01 AI works, copyright is not clear
Open Source China: First, let’s clarify the concept. Are copyright and authorship the same thing?
Deng Chao: Yes, there is no difference between copyright and copyright. According to relevant legal provisions, copyright refers to the rights of publication, signature, reproduction, and the right to protect the integrity of the work that the author of the work and other relevant personnel enjoy in accordance with the law. And copyright is copyright.
Open Source China: OK, according to the current law, do individuals or organizations own the copyright for images produced using AI ?
Deng Chao: In the current law, this area has not been clearly stipulated, and industry opinions are not unified. Case-by-case analysis is required.
The year before last, there was a case of "Film v. Baidu" in the Beijing Internet Court. At that time, the court held that machine-generated content has no copyright.
However, in the Shenzhen "Tencent News Case", the court held that machine-generated content has copyright.
The South and the North have given different answers to this question, but these cases have not been appealed, so there is no unified opinion yet. In today’s case, the Beijing Internet Court held that images generated by AI are copyrightable. However, the Beijing Internet Court is a grassroots court, and the effectiveness of its judgments is limited. It is unclear whether it can be promoted nationwide.
Open Source China: Are the rights of AI -generated images very vague?
Deng Chao: It can’t be said to be very vague. It involves a process of judgment. The judgment also emphasized that whether content generated by artificial intelligence constitutes a work depends on the circumstances of the case and cannot be generalized. In this case, the generation of this picture involved hundreds of prompt words and parameter adjustments. In this case, the court will recognize that the creator has put in a certain amount of creative work. On the contrary, if you just tell the AI to "draw an autumn landscape", then in my opinion this cannot be copyrighted, because there is no creation that meets the requirements of copyright law.
Open Source China: However, for these parameters and settings output for creation, how much is too much and how much is too little?
Deng Chao: Yes, this is exactly the difficulty of judgment. In every case, every judge's judgment may be different.
Open Source China: Some lawyers believe that AI drawing is very different from traditional art creation. The plaintiff only output prompt words and did not draw specific lines, nor did it tell the Stable Diffusion model 100% how to draw specific lines. The lines and colors are used to generate pictures through artificial intelligence. According to the traditional theory of the Copyright Law, only works created by human beings are protected by copyright, and the actions of the plaintiff Li Yunkai did not constitute creation. What do you think?
Deng Chao: Yes, this objection is quite common. To make an analogy, I ask a painter to paint an autumn landscape, but I cannot say "I am the author of this painting" after the painter has finished painting. This is unreasonable.
There is another objection, and its logic is: Even if I say "draw an autumn landscape", different people will draw different things. It cannot be because I said this instruction, I It’s very strange to own the copyright of a painting. Things like autumn scenery can be painted in a thousand different ways by a thousand people. Even if it is an AI, different AIs can draw it differently. Even if you ask the same AI to draw it again, it may not be able to draw it.
Open Source China: Even with the same parameters, it does not necessarily generate the same graph.
Deng Chao: Yes, that’s right. But this analogy can only be applied to humans. It is different when applied to AI, because AI is a machine, not a specific person, and has no rights. If I asked one person to draw this picture, then the copyright must belong to the person who drew it. But AI is not a human being, and it cannot be an author. At this stage, generative artificial intelligence models do not have free will, and neither domestic nor foreign countries support AI becoming a legal subject.
This is also the consideration made by the court in this judgment. At present, whether you think AI works should have copyright or not, it can actually be explained, because AI is a new thing and the law is relatively backward. There is no consensus yet on what should be done in this situation? Therefore, when the court makes a ruling now, it is more from the perspective of industry or legislative purposes to explain: Is it better to protect this AI, or is it better not to protect it?
In this case, in order to generate this picture, the plaintiff entered a large number of instructions and adjusted hundreds of parameters. After a series of forward and reverse constraints, this picture was obtained. This process, in the view of the court, is somewhat original. The originality of our country's legal provisions is very low. For example, in photography, if the photographer waited for a day and arranged for a specific angle to take the picture, everyone will think it is original; but the pictures we ordinary people take casually are generally original, even if you have not adjusted the parameters. No ideas either. Therefore, the threshold for originality is very low, and you don’t need to reach any professional level to possess it.
Open Source China: But we took the photos ourselves, not someone else told me how to take the photos. And when using AI , no matter how specific the description is, it is only text creation, not art creation. It cannot predict or control the results of the creation. This is very different from the way people used to use brushes and drawing software to draw pictures. . It is a "Vincent picture", not a "hand-made picture".
Deng Chao: So, this is actually a question of value orientation, and it makes sense whether to provide copyright protection or not.
There is an objection that, for example, the instruction "draw an autumn landscape" can have textual copyright, but this textual copyright cannot be extended to the pictures it generates, otherwise all future autumn landscape paintings will infringe My copyright has been lost because I was the first to say this instruction.
However, from a historical perspective, it is a matter of time before all forms of creation can obtain copyright protection. For example, when photography technology was first born, everyone felt that photography should not be copyrighted, because before the advent of photography, everyone used oil paintings, and it might take decades of study to paint landscapes that look very similar. As a result, after the advent of photography technology, you don’t need any painting skills or decades of study. Just click the shutter and you can get a painting that is closer to reality than if you studied painting for more than ten years, so everyone thinks it is It's not art and shouldn't be copyrighted. But later you see, with the development of technology, photography has slowly become an art.
From a historical perspective, whenever something new appears, it will always face some controversy, such as whether to grant copyright? But we can find ways to better protect copyright and adapt to the emergence of new things while protecting rights.
02 Can existing laws be referred to in new situations?
Open Source China: There is now an idea to compare AI creation to Internet products. For example, a programmer enters code in the background and generates an APP through the computer with its various functions. But code is not a work of art. It doesn’t matter who writes it, as long as it can be implemented. The code behind the same function may be exactly the same. In AI creation, the prompt word is equivalent to the "code" of creation, and the generated picture is similar to the front-end APP. So, is the prompt word the same regardless of who writes it? Should there still be copyright?
Deng Chao: In the field of intellectual creation, the threshold for legal requirements is very low. Picasso’s paintings have copyright, and so do elementary school students’ paintings. It is not measured by the level of intellectual contribution. As long as you write it, you have copyright.
Therefore, code can also enjoy copyright, such as software copyright registration, which registers the code. If you write a code and someone else plagiarizes your code, that also constitutes copyright infringement. However, the copyright of the code is limited to the text code itself.
Initially, it was controversial to protect code as literary fiction. After all, text does not have any practical functions, but code, its only purpose is to implement functions. However, copyright law does not protect functions. If you want to protect functions, you can use patents. Therefore, there is such an inconsistency in treating codes as literary novels at first. Then the law later stipulated that unique or very limited code that implements a certain function cannot be protected by copyright. For example, writing a "pop-up prompt box" alone cannot be protected by copyright. But if the entire program, such as WeChat, as a whole, it is protected by copyright.
If we take music as an analogy, a single syllable such as do, rei, mi, fa, and so cannot be protected by itself, but if you arrange it into a piece of music, then the piece of music can be protected by copyright.
Open Source China: This means that it must be presented as a whole work to be considered.
Deng Chao: Yes. But to what extent it should be presented, there is actually no clear line. For example, if a song counts, does one section of it count? This requires detailed analysis in specific cases.
Open Source China: Artificial intelligence software can be copied on a large scale. As long as the computing power is sufficient, there is no problem in generating a large number of images in a short time. If all this can be protected by copyright, will it not have the effect of encouraging innovation for the entire society?
Deng Chao: That’s right. From a practical point of view, it is impossible to tell whether many pictures were generated by AI. Except for those that are poorly hand-drawn, it is actually very difficult to tell. In this case, we just assume that all pictures are copyrighted. Don't care whether they are produced by AI. This has nothing to do with AI. When we use pictures, we follow two principles: if you are original, or if you purchased the license legally, then use it; if not, then don’t use it casually. You can't just pick up a picture when you see it online. Regardless of whether it is produced by AI or not, the risk is very high.
If the picture I generated using the AI large model is the same as that generated by others, for example, under the same prompt words, the AI generated the same picture for us, then it doesn’t matter, because you created it independently, and you can avoid legal action. Risk, even if the other party comes to sue, you are not afraid.
Open Source China: Does this mean that we each have copyrights for the two pictures generated by the same AI and the same set of prompt words, even if the two pictures are exactly the same?
Deng Chao: Yes, because we are talking about originality. As long as you complete it independently, there is no problem. For example, if two people took pictures of the flag raising in Tiananmen Square at the same time, their angles may be very similar, and the pictures they took are almost the same. In this case, each has a copyright and no one has infringed the copyright. But if it really comes to the court stage, you have to provide evidence to prove that this picture was indeed taken by me and not by me. In this case, the defendant cut off the watermarks of the plaintiff's pictures, which must have been taken from the Internet and must be infringing. But if the defendant can show the original picture or prompt words to prove that the picture can indeed be generated, then the defendant's infringement will not be established.
Open Source China: But if we want to reproduce it, if we enter the same prompt word and the AI- generated picture is different, then we won’t be able to prove it ourselves?
Deng Chao: This is theoretically true, but on the other hand, in these picture infringement cases, the amount of damages awarded is very low. In this case, the amount of damages awarded is 500, and in other cases it is even less, usually only a few dozen yuan, and the risk is very small. If it is a self-created picture, although the creative process may not be preserved, it will definitely leave traces. If this is not possible, witnesses can be found to testify in court. In short, it will definitely be found out, so it will not have any subversive impact on the status quo.
03 AI creation is extremely risky
Open Source China: Speaking of originality, many current AI pictures are "dissolved" from other people's existing pictures. Can this also be considered originality? Doesn't this count as plagiarism?
Deng Chao: There is currently a case: On December 27, 2023, the "New York Times" sued Microsoft and OpenAI, claiming that millions of newspaper articles were used as AI training data. Is the feeding of these data considered fair use or openAI? Copyright infringement has not yet been determined, so you can pay close attention to it.
The "fair use doctrine" means that although your actions are technically infringement, your actions are an acceptable borrowing to promote creative expression. For example, scholars can quote and excerpt others' content in their own works; authors can publish adapted books; ordinary people can intercept movie clips to make film reviews. In other words, if copyright restrictions are too strict, civilization's creativity may stagnate.
Technology companies have long used this principle to circumvent copyright disputes. In 2013, Google was sued by the Authors Guild for copying millions of books and uploading excerpts from them online. A judge ruled that Google's behavior was legal based on the fair use doctrine because it created a searchable index for the public. Created public value. In the age of big models, the fair use doctrine may still play a key role. People who support AI non-infringement believe that the process of large-scale model generation of content is similar to human creation - when you try to draw a painting or shoot a video, you will also have the paintings or movies you have seen in your mind. Human creations progress on the basis of predecessors, and the same goes for large models.
Reality is always one step ahead of the law: when problems arise in reality, the law responds, but now it just happens that there is no consensus.
In essence, the purpose of copyright law and intellectual property law is to promote social prosperity. In ancient times, there were no intellectual products. Before the advent of printing in the Song Dynasty, everyone used hand-written manuscripts, but everyone thought this was a good thing and could prevent documents from being lost. After the advent of printing, copyright gradually became available. Legislators believe that providing certain protection to authors of works can inspire others to create more works. The reality is indeed like this. Developed countries all have very complete knowledge accumulation. So in essence, it still depends on which choice can better promote the development of the industry and the entire society.
Open Source China: When it comes to promoting creation, I want to talk about it here. In the art world, if the similarity reaches 10%, it is considered plagiarism. However, AI is now plagiarizing without a statement and not using it at any cost. Isn’t this right? Is it exactly the opposite of the original intention of promoting creation? If it will be plagiarized as soon as it is made, wouldn't it just dampen people's enthusiasm for creation and squeeze people's creative space?
Deng Chao: Yes, that’s right. For everything, the impact of policy has two sides, and you are talking about the negative side. But on the plus side, creating pictures in the past required artists to study art for ten years, and it took an artist a day to paint a picture. But now, anyone who has never studied art can create a painting. From this aspect, productivity has definitely been greatly improved. The pictures generated by AI in one day can equal the creations of 100 artists. Of course, fine-tuning is required in the end, but productivity has been doubled.
Assume that a company now needs 100 pictures. If it hires an artist, the cost will be 100,000 yuan; but if it hires an AI, it may only cost 1,000 yuan. Including later manual modification, it may only cost 10,000 yuan. From the company's perspective, its costs have been reduced and it has become more competitive, but ten artists behind the scenes may have lost their jobs. But from another perspective, there may be ten more AI-related jobs added.
So AI is like the butterfly effect. The social changes brought about by a change in one factor are very complex. Some people will benefit and others will suffer. So it is difficult to predict here, and it is not something that I can figure out alone.
Open Source China: In terms of risk, here is an example: Suppose I only input text instructions when using AI to make pictures , but the AI was fed copyrighted pictures during training, so the pictures produced for me also infringed. Well, am I responsible or the AI company is responsible?
Deng Chao: According to the case we are discussing, you must be responsible, because AI is just a tool, just like a paintbrush or a camera, and someone may still be responsible for problems in the end. In addition to the "New York Times v. OpenAI case" we just mentioned, because OpenAI used the content of the New York Times during training, OpenAI is responsible. But now that OpenAI has provided product software, if the pictures you generate with this AI software infringe (although there is no case yet), the AI tool will most likely not be responsible, and the responsibility lies with you.
Open Source China: But I didn’t intend to infringe anyone, nor did I ask this AI to plagiarize other people’s works. I didn’t even know the existence of other people’s works. It was the AI company that fed copyrighted images during training to generate the works. Infringement, I have no control over this situation, do I need to be responsible?
Deng Chao: Indeed, the three questions you asked are the three most difficult questions in the AI era, and no one has the answers yet:
First, large models use massive amounts of data for training. Is this behavior infringement or fair use?
Second, do AI- generated works have copyright? (In the case discussed today, the Beijing Internet Court has already given its preliminary opinion)
Third, who will be held responsible if works generated using AI infringe?
No one has the answers to these three questions now. What I just said is just my personal opinion.
Open Source China: That means that if there are benefits, the users will get them, and if there are risks, they will be borne by the users. Then isn’t the AI company only responsible for making money and not taking risks?
Deng Chao: In simple terms, it is like this, but the details must be combined with the specific circumstances of the case.
The risk for enterprises is the first question we asked: Is using other people’s content to train AI models considered fair use? The policy at this time is very sensitive. Assuming that this is considered a reasonable use and the corporate risk is very small, it will promote the development of the AI industry to a certain extent, because the enterprise will have no worries; but if the enterprise is allowed to bear a large responsibility, then some small companies that have just developed may not be able to They will go bankrupt due to legal proceedings, which will hinder the development of the industry. Whether it is law or judgment, industrial development must be considered.
Open Source China: But we can’t just reap the benefits and leave the risk to users to argue with the original author, right?
Deng Chao: Yes, but if someone benefits, someone will definitely suffer. This is unlikely to be a win-win situation. It depends on whose value orientation you prefer. For example, when the industry is in its early stages, it may be inclined to promote industry development; when the industry matures and the company grows, it would not be good if it does not take any responsibility at all. Industrial policies are different at different stages. In terms of law, at least in my opinion, both referees and legislators need to be balanced.
Open Source China: With the development of artificial intelligence and large language models, what other legal issues may be involved in the production method of AI creation?
Deng Chao: In addition to the three points mentioned above, there is also the possibility of infringing on other people's portrait rights . For example, if the generated portrait looks the same as someone else, it may infringe on the portrait rights. This is also a type of infringement in the third point. Form it.
Open Source China: How does the legal community view AI creation, a new production model? Will relevant laws and regulations be introduced next?
Deng Chao: Judging from past historical experience, the law will not be introduced so early. In the past, for new things like the Internet, the legal community would let the bullets fly for a while, observe them for a while, and then promulgate relevant laws and regulations when the time was right. In addition, our current intellectual property law framework is sufficient to solve some of the current problems related to AI. There is no need to legislate a separate law for AI. Maybe when the time comes, some explanations or regulations under the law will be improved.
What do you think about this? Do you think AI works should have copyright? See you in the comment section~
Guest of this issue: Deng Chao | Lawyer (WeChat ID: dengchao)
He has a bachelor's degree in science and a doctorate in law, and has a profound legal theoretical foundation; he has been working in the intellectual property industry for more than ten years and has rich practical experience.
He has long been focused on the research and practice of cutting-edge intellectual property issues in the fields of technology and media, and represents clients in handling related litigation and non-litigation legal matters.
Before entering a law firm, he worked in the legal department of a Fortune 500 company and a top domestic intellectual property firm, providing intellectual property legal services to many multinational companies and domestic listed companies.
Fellow chicken "open sourced" deepin-IDE and finally achieved bootstrapping! Good guy, Tencent has really turned Switch into a "thinking learning machine" Tencent Cloud's April 8 failure review and situation explanation RustDesk remote desktop startup reconstruction Web client WeChat's open source terminal database based on SQLite WCDB ushered in a major upgrade TIOBE April list: PHP fell to an all-time low, Fabrice Bellard, the father of FFmpeg, released the audio compression tool TSAC , Google released a large code model, CodeGemma , is it going to kill you? It’s so good that it’s open source - open source picture & poster editor tool