Many game companies have introduced AIGC tools into their art design pipelines in order to improve efficiency in game updates and iterations. However, the use of AI-generated images often gives rise to a series of new legal questions, including but not limited to:
Can art assets created using AIGC platforms be commercially exploited?
Can rights be asserted over art assets created with the assistance of AI?
Does “running content through AI once” automatically eliminate infringement risks?
Can prompts published online by others be used directly?
...
Against this backdrop, this article analyzes the issues from three perspectives that are particularly relevant to game companies: the ownership of AIGC-generated art assets, infringement risk analysis, and infringement risk mitigation.
Ownership of AIGC-Generated Art Assets
At present, the question of copyright ownership in AI-generated content remains highly controversial in both academic discourse and judicial practice. Existing laws and regulations do not provide a definitive answer. Among effective court judgments, some recognize human authors as copyright holders of AI-generated images, while others reach the opposite conclusion. Representative cases are summarized below.
In the first “text-to-image AI” case decided by the Beijing Internet Court—the “Spring Breeze Brings Tenderness” case—the court held that the image constituted a work of fine art reflecting the plaintiff’s individualized expression. The plaintiff’s creative process consisted of three stages: (i) selecting an open-source model, (ii) selecting prompts and negative prompts, and (iii) adjusting relevant parameters. Based on these facts, the court found that the plaintiff had made intellectual contributions in terms of character presentation, prompt selection and sequencing, parameter configuration, and selection of outputs. Accordingly, the plaintiff was held to enjoy copyright in the image.
Similarly, the Shanhaiguan District People’s Court of Qinhuangdao City, Hebei Province, held that although the image was AI-generated, it reflected the author’s unique choices and arrangements in parameter adjustment, style design, and keyword selection, and therefore should be protected under the Copyright Law.
The Tiefeng District People’s Court of Qiqihar City, Heilongjiang Province, held that the plaintiff’s refined language instructions, adjustments to lighting and color layering, and post-generation selection and modification resulted in a personalized outcome distinguishable from random machine output, thereby qualifying for copyright protection.
The Donghu High-Tech Development Zone People’s Court of Wuhan City, Hubei Province, found that the correspondence between keywords and visual elements demonstrated a degree of “mapping,” and that the creator exercised a certain level of “control and foreseeability” over the generated work, which warranted protection.
By contrast, in the “Phantom Wing Transparent Art Chair” case decided by the Zhangjiagang People’s Court of Suzhou City, Jiangsu Province, the court acknowledged that AI-generated content may in principle be copyrightable. However, because the plaintiff had only input a single, simple set of prompts and parameters, lacked decisive influence over the output, and could not reproduce the generation process, the court found insufficient originality and denied copyright protection. The court emphasized that a key criterion lies in whether the user can provide original records of the creation process proving that, through prompt refinement and parameter modification, the user made personalized and substantive contributions to expressive elements such as layout, proportion, perspective, composition, color, or linework.
A consistent conclusion emerges from existing case law: judicial practice generally affirms the potential copyrightability of AI-generated content. However, whether AI-assisted images created by human users are protected depends on whether they satisfy the standards of “independence” and “creativity.”
“Independence” refers to whether the image is primarily attributable to human input—i.e., whether the human author exercised decisive influence over the output.
“Creativity” refers to whether the work reflects intellectual input and individualized choice by the human author.
Accordingly, if a game company wishes to effectively enforce rights against third-party copying of AI-generated art assets, it should proactively retain evidence of authorship from the following perspectives:
1. Use multiple rounds of detailed prompts and negative prompts, covering artistic themes, styles, subjects, decorations, environments, lighting, composition, color schemes, and other expressive elements.
2. Adjust generation parameters (e.g., image ratio, reference values, guidance coefficients), and combine AI output with tools such as Photoshop and manual drawing to increase human intellectual contribution.
3. Attempt reproducibility, by using the same model, random seed, prompts, and parameters, to demonstrate control over the output.
4. Record the entire creation and reproduction process, and consider blockchain-based evidence preservation to enhance evidentiary weight.
Until clear legislation, judicial interpretations, or guiding cases are issued, no AIGC art asset can be guaranteed copyright protection. However, the more specific the prompts, the greater the degree of human post-processing, and the higher the reproducibility of results, the greater the likelihood of protection. Game companies are therefore advised to focus their efforts accordingly when completing art design.
Infringement Risk Analysis of AIGC Art Assets
Many game companies use existing images as base or reference materials in “image-to-image” generation, or directly adopt pre-existing base models available on AI platforms. Whether AI-generated images can be directly used as in-game art assets or extended into derivative merchandise depends on whether the base model is open-source, whether the platform has obtained authorization from non-open-source model creators, and the platform’s specific commercial use policies.
In the Ultraman Model Infringement Case, users introduced an Ultraman LoRA model into an AI platform, allowing others to generate similar images. Although the Hangzhou Internet Court’s judgment ((2024) Zhe 0192 Min Chu No. 1587) did not directly characterize the users’ conduct, it held that the platform infringed the right of information network dissemination by providing online browsing, downloading, and sharing of unauthorized content.
More recently, in the Medusa case decided by the Jinshan District People’s Court of Shanghai, the court explicitly held that training a LoRA model using existing IP assets and making image sets and videos publicly available infringed the copyright in the Medusa character.
These cases suggest that where an AI base model is not open-source, or involves a well-known IP, its use may constitute copyright infringement or unfair competition.
Accordingly, infringement risks for game companies primarily arise in the following scenarios:
1. Direct Use of Public Models or Reference Materials on AI Platforms
Most AIGC platforms specify IP ownership rules in their User Agreements and may further publish Commercial Use Guidelines indicating permissible use scopes. Where game companies comply with such disclosed rules, infringement risks are relatively low. In the absence of clear commercial policies, infringement assessment depends on substantive similarity analysis between the base model and the generated content.
2. Using Existing IPs as Training Materials
Even if IP models are labeled as “open-source” or “commercially usable,” mislabeling or inadequate platform review may exist. For highly well-known IPs, courts generally impose an obligation of avoidance. If generated content is substantially similar to existing IPs, infringement or unfair competition risks remain.
3. Using Publicly Available Prompts
The first AI prompt copyright case decided by the Huangpu District People’s Court of Shanghai held that prompts consisting of unordered keyword lists lacked originality and did not constitute copyrightable works.
In general:
Simple keyword combinations pose low infringement risk.
Coherent but purely functional descriptions may constitute limited expression and remain unprotected.
Prompts derived from literary works may involve infringement risks.
Moreover, prompts are typically unobservable from final outputs, and infringement assessment ultimately depends on whether the generated content is substantially similar to protected works.
Mitigating Infringement Risks of AIGC Art Assets
To mitigate risks, game companies are advised to prioritize proprietary assets and public domain materials. Where third-party assets are necessary, companies should obtain licenses or use materials explicitly authorized for commercial use, followed by IP risk assessment prior to in-game implementation.
Use of Proprietary Assets:
Ensure the company holds complete IP rights and that commissioned works do not reserve AI training rights.Use of Public Domain Materials:
Avoid confusion between public domain works and protected derivative works.Use of Third-Party Assets:
Obtain authorization where substantial similarity is unavoidable, retain training materials, and conduct comparative analysis to assess infringement risks.
Conclusion
Before definitive legislation or guiding cases emerge, no AIGC art asset can be guaranteed copyright protection. However, existing judgments clearly focus on one core issue: demonstrating human creative contribution. Multi-round prompt design, precise parameter control, substantial manual modification, and comprehensive evidence preservation all increase the likelihood of protection.
AIGC is a powerful tool—but not a liability shield. Game companies should pursue efficiency while maintaining clear boundaries, conducting prior IP assessments, and avoiding unnecessary legal and reputational risks.
