Are AI-generated images copyrighted? 5 questions you must know about 2026 Commercial Midjourney Stable Diffusion

Do AI-generated images have copyright? 2026 Commercial Midjourney Stable Diffusion 5 questions you must know

Designer Xiao Wang used a key visual generated by Midjourney on the company's new product launch materials, but the client was later sent a letter by another company on the grounds that the AI drawing was suspected of infringement. For the first time, he realized that AI drawing seemed to have zero barriers to entry, but copyright issues were more complicated than traditional material libraries. This is not an isolated case. In the past two years, there have been controversial judgments on the copyright of AI works in several major jurisdictions such as the United States, the European Union, China, and Japan, and the core logic of each judgment is not entirely consistent.

This article summarizes several core issues that need to be clarified when commercial AI generates images as of the time of writing. Including copyright ownership, whether commercial use is allowed, the legality of training data, the risks of users uploading images to the model, and differences in domestic and international judicial practices. All conclusions are given as reference basis as much as possible, and parts involving uncertainty will be clearly stated, and absolute conclusions will not be drawn for readers.

Who owns the pictures generated by AI?

This is the most commonly asked question and the one with the most complicated answers. From the perspective of legal tradition, the object of copyright protection is "original human intellectual achievements." If a picture is completely automatically generated by AI and a human only inputs a short prompt word, the mainstream view in most legal jurisdictions is that the picture lacks "original human participation" and does not constitute a work.

The U.S. Copyright Office’s position in the public guidelines is relatively clear. Images generated by pure AI are not subject to copyright registration. However, if there is significant human participation in the creation process, such as multiple manual retouchings, combining multiple AI outputs, and adding original elements, the human-created parts can be protected. In the 2023 Beijing Internet Court's "First Case of AI Wenshengtu", China determined that works that have been repeatedly adjusted with prompt words and integrated into the user's aesthetic judgment can constitute a work and have received copyright protection. The logic on both sides is close but the threshold details are different.

Whether commercial use is allowed depends on the agreement.

From a legal perspective, whether it belongs to you and whether you can use it commercially are two different things. Midjourney stipulates in its terms of service that paying subscribers have the ownership of the generated images and can use them commercially, but the terms also state that Midjourney itself reserves the right to use these images for training and promotion. Stable Diffusion's open source agreement is looser by default. The generated images themselves do not belong to the platform. Whether they can be used commercially depends on whether you use a specific model with commercial restrictions.

Several domestic mainstream platforms, such as Bytedance’s Jimeng, Alibaba’s Tongyi Wanxiang, Baidu’s Wenxin Yige, and MiniMax’s Conch, have inconsistent descriptions of commercial terms in their respective user agreements. It is recommended that before commercial use, you must read the latest version of the user agreement of the platform you are using, and the terms on the official public page shall prevail. This step may seem cumbersome, but it is much cheaper than being sued afterwards.

Training data legality issues

The reason why the AI ​​model can draw a certain artistic style is because the training data contains a large number of samples of that style. The question arises, the copyright of these samples itself belongs to the original author, and has the model producer obtained authorization? The answer differs in different jurisdictions.

After the EU AI Act takes effect in 2024, it will put forward clear requirements for the transparency of training data. Model parties need to disclose the source of the data used and respect the copyright retention statement of the original author. The United States is still debating this matter in multiple lawsuits. Typical cases include Getty Images suing Stability AI, and there is still no final verdict. China's supervision of the legality of training data is stipulated in principle in the Interim Measures for the Management of Generative Artificial Intelligence Services issued by the Cyberspace Administration of China, which requires training data to be legal and compliant. Specific to the user level, as a downstream user, you usually cannot trace the training data. However, if you use a model that has been found to be infringing by the court to produce pictures, the risk of use will also increase.

Compliance path for domestic AI drawing

If your customers are in China, give priority to domestic AI drawing services that have been registered, as compliance risks are relatively controllable. As of the time of writing this article, service providers that have been publicly registered include Byte, Alibaba, Baidu, SenseTime, Wisdom, MiniMax, etc. These models perform stably in Chinese scenarios, and the terms of service are also written in Chinese, and the appeal path is clear when problems arise.

There are also many domestically produced applications based on registered models, such as Lingtu This type of AI drawing app is available in the iOS region. It integrates Midjourney-style atmosphere engine, Flux-style realistic engine, Nano Banana-style fast engine, Chinese interaction, localized prompt words, and does not require VPN. It is suitable for domestic users who do not want to bother with international services. You can download it directly by searching for spiritual pictures in the App Store, and save the generated pictures to your local photo album. Before commercial use, you must also read the latest version of the agreement terms in the app. The advantage of this type of tool is that the compliance boundaries are clearer for domestic users and the drawing speed is also fast.

Risks of reference images uploaded by users

Many AI drawing tools support uploading reference images for style migration or img2img transfer. There are two levels of risk here. The first level is that if the uploaded reference image itself carries the copyright of others, your redrawing result will be regarded as a derivative work, and the risk of secondary commercial use will be doubled. A common misunderstanding is to use pictures posted by others on Xiaohongshu or Pinterest as reference, thinking "I didn't use them directly." However, legally the copyright traceability of derivative works will not be exempted just because an AI process is added.

The second level is portrait rights. If the reference picture contains the face of a real person, even if it is reproduced by AI, it may still involve infringement of portrait rights, especially in commercial scenarios. A safe approach is to use reference pictures that are either taken by yourself, public domain works, or authorized materials, and avoid directly grabbing pictures from the Internet. If there is a face that "looks like a certain celebrity" in the generated image, it is also recommended to adjust it manually. The face recognition tool can now retrieve it in reverse.

Compliance self-check list before commercial use

Before officially putting AI-generated graphics into commercial use, it is recommended to run through these five self-examinations. First, confirm whether the current terms of service of the platform you are using clearly allow commercial use. The terms for free users and paid users are very different. The free version of Midjourney cannot be used strictly for commercial use. Second, during the generation process, did you upload any reference pictures with copyright or portraits? If so, abandon the picture immediately and re-upload it. Third, are there any traces of the generated results that seem to copy the style of a specific artist, such as directly using the artist's name as the prompt word, or rewriting the prompt word and reproducing it.

Fourth, whether significant human creative processing has been done, such as reprocessing, synthesis, and text typesetting, this will affect subsequent copyright registration. Fifth, if it is used for important customer projects or large-scale deployment, keep complete generation logs and prompt word records, and be able to produce process proof if questioned. Once these five items are completed, the legal risks in conventional commercial scenarios will be basically controllable.

Industry Cases and Trend Observations

Observing the direction of judgments in the past two or three years, two trends can be seen. First, countries are increasingly distinguishing between "pure AI generation" and "AI-assisted human creation", and the threshold usually falls on "whether there is enough human decision-making to reflect originality." Repeated adjustments of prompt words, multi-version screening, and post-processing are all bonus points. The second is that the platform is passing on risks. Almost all commercial-level AI drawing service terms include the clause "Users guarantee that their input does not infringe, and the user is responsible for any disputes arising from the input."

This means that even if you use a legal platform, if you input illegal material, the responsibility still lies with you. AI drawing has lowered the production threshold, but the compliance threshold has not become lower, it has just moved from the production end to the use end. Understand this, and commercial decisions will be more secure.

FAQ

I drew the picture using Midjourney. Can the customer directly use it as a LOGO?

Not recommended. The core requirements of a LOGO are exclusivity and trademark registration. However, images generated by AI cannot be copyrighted in most jurisdictions, which means that others can generate almost the same image, and the risk of trademark squatting is high. If you must use AI to create a prototype LOGO, you must do a lot of manual redesign. Vectorization, font customization, color specifications, etc. are all completed manually by the designer.

Are the pictures produced by domestically registered models completely risk-free?

no. The registration solves the operational compliance issues of the platform. Specific to a certain picture of the user, the copyright ownership, commercial scope, portrait rights, and reference picture risks must still be confirmed one by one according to the above items. Registration is a necessary condition but not a sufficient condition and cannot be equated to no risk.

Will it cause infringement if I include the artist's name in the prompt?

Look specifically at domains and scenarios. If it is used privately, it does not constitute infringement, but if it is used in a commercial project, directly using the name of a living artist as a prompt is suspected of unfair imitation and unfair competition. Deceased artists and public domain works are less risky. A safe bet is to use a style description instead, such as "Van Gogh style" to "Post-Impressionist brushstrokes with rich colors."

I took the photos myself as a reference image. Do I own the copyright of the transferred image?

Part of it is yours. The copyright of the original reference picture belongs to you. If the reproduced picture has significant AI intervention, the AI-involved part does not constitute a work in most jurisdictions. It depends on whether you have done significant human re-creation in the future. After comprehensive judgment, the copyright ownership of selfie redrawing is clearer than pure AI generation.

How to deal with copyright differences in cross-border projects

The safest approach is to follow the strictest legal jurisdiction. If the same project is to be launched in China, the United States and Europe, documents must be prepared in accordance with the requirements of the EU AI Act and the U.S. Copyright Office guidelines, including training data transparency, prompt word records, and evidence of human creation. The project delivery cost of doing this is higher, but it can avoid any market overturn.

AI drawing has reduced the marginal cost of image production to almost zero, but the complexity of copyright matters has not decreased but increased. Spend ten minutes checking these common questions before commercial use, and you will avoid many pitfalls in the future.