Rights Issues with AI-Generated Images: Copyright, Portrait Rights, Trademarks, Design Rights, Similarity, and Provenance

About 10 minutes

Corporate legal and IP staff using generative AI for image creation, creators, and AI product developers

AI and Copyright: Comparing Legal Frameworks in Japan and the US

The proliferation of AI image generation technology has exposed creators and enterprises to complex legal risks spanning copyright, portrait rights, trademark law, and design rights [1][2][6][8][10]. This article organizes how each of these rights applies to generated images and what enterprises should do in response.

Copyright: Protection and Infringement Risk for AI-Generated Images

Copyright Protection of AI-Generated Images Themselves

Under the current Japanese interpretation, copyright requires a creative expression of thoughts or feelings [1]. The Agency for Cultural Affairs’ view on AI and copyright also explains that simple prompt input may involve little creative contribution, while copyright may arise in portions where human creativity is exercised through detailed prompt design, post-generation editing, or selection and curation of AI output [2].

Under the U.S. position (U.S. Copyright Office, 2023 guidance), the Zarya of the Dawn decision clarified that AI-generated portions do not receive copyright protection, while only the portions created by humans are protected [3][4].

Perspective	Japan	United States
Protection of AI-generated images	Depends on the degree of human creative contribution [1][2]	AI-generated portions unprotected; human-created portions protected [3][4]
Clarity of standard	Interpretive room (no explicit statutory provision) [1][2]	Certain guidance available from Copyright Office [3]

Similarity and Provenance: The Two Requirements for an Infringement Determination

Provenance (access + intent to reproduce) Generative AI may ingest large quantities of copyrighted works as training data. If a model is strongly influenced by a specific work and generates output that reproduces its expression, provenance may be found [2][5].

Similarity (substantial similarity) Whether the output image is “substantially similar” to the original work is assessed based on the degree of correspondence in composition, color, style, and individual expressive elements. Note that style and artistic manner are themselves not protected by copyright, so a prompt for “an image in the style of [artist]” does not immediately constitute infringement. However, output that faithfully reproduces the expression of a specific work may be problematic [1][2].

Provenance + Similarity     → High copyright infringement risk
No provenance (independent creation) → No infringement even if similar
Provenance + Low similarity  → Low infringement risk (depends on specifics)

Memorization Risk

Research has confirmed that some generative AI models “memorize” specific content from training data and output it in nearly identical form (memorization) [5]. This represents the most direct risk of unintentionally reproducing a third party’s copyrighted work almost exactly.

Portrait Rights: Risks of Generating Real People’s Likenesses

The Structure of Portrait Rights

Portrait rights are not explicitly codified in statute, but are established by case law as a right with two main dimensions [6]:

Portrait rights as privacy (including private individuals): The right not to have one’s appearance or figure photographed, published, or used without consent
Publicity rights (celebrities): Rights over the commercial exploitation of the customer-drawing power of a celebrity’s name and likeness

Specific Risks in AI Generation

Risk Scenario	Rights at Issue
Using an image closely resembling a celebrity’s face in an advertisement	Publicity rights infringement
Generating and distributing sexual or defamatory images of a real person	Portrait rights infringement, defamation, privacy violation [6][7]
Generating and publishing identifiable images of private individuals without consent	Privacy rights violation
Commercially exploiting a deceased celebrity’s likeness	Publicity rights (may persist for a certain period after death)
Creating deepfake disinformation	Defamation, obstruction of business by fraudulent means, and others [7]

Key point: Even when AI generates a “similar-looking person,” portrait rights issues may arise if the output is sufficiently specific to identify a real individual. Using a celebrity’s name in a prompt may be treated as a deliberate attempt to generate that person’s likeness.

Legal Responses to Deepfakes

Japan currently lacks comprehensive legislation directly regulating deepfakes, but individual laws and ordinances, including the Criminal Code provision on defamation (Article 230), may become relevant [7].

Trademark Rights: Brand Logos and Trade Dress

Scope of Trademark Rights

Trademark rights prohibit the use of a mark identical or similar to a registered trademark in connection with the same or similar categories of goods and services [8]. The main scenarios in which AI-generated images pose problems are:

Direct logo reproduction Cases where a specific brand’s logo is specified in a prompt, or where a logo reproduced from training data appears in the output.

Imitation of trade dress (distinctive appearance) Product packaging color schemes, store design, and character appearances may be protected as “commercial representations” under the Unfair Competition Prevention Act even when not trademark-registered [9].

Similarity to characters Generating images closely resembling well-known characters for commercial use may give rise to trademark infringement, copyright infringement, or both [1][8].

Relationship Between Trademark and Copyright Infringement

Reproducing an identical logo in an AI image → Potential copyright + trademark infringement
Selling a product bearing that logo image  → Often processed as trademark infringement
Non-commercial criticism or parody of a logo → Trademark may not apply (different standard from copyright fair use)

Design Rights: Visual Appearance of Products, Buildings, and UI

What Are Design Rights?

Design rights protect the aesthetic appearance of articles, buildings, and images—specifically their shape, pattern, color, or combinations thereof that create a visual aesthetic impression (Design Act, Article 2) [10]. Unlike copyright, design rights require application and registration with the Japan Patent Office to take effect [10].

Under the Design Act after the 2019 amendment, articles, buildings, and images are included within the scope of protectable designs [10].

Protected Subject Matter	Examples
Product designs	Shape and design of appliances, furniture, packaging containers
Architectural designs	Building exteriors and interior design of commercial spaces
Graphic designs	UI design and graphics on operating screens

How Design Rights Apply to AI-Generated Images

Design rights become an issue primarily when AI-generated images are used for product design proposals and prototyping.

Risk Scenario	Rights at Issue
AI-generated product design closely resembles a registered design	Design rights infringement (when commercially implemented) [10]
AI-generated building exterior or interior resembles a registered design	Design rights infringement [10]
AI-generated UI or graphic resembles a registered graphical design	Design rights infringement [10]
Competitor’s product design specified in a prompt and then commercialized	Design rights infringement + Unfair Competition Prevention Act [9][10]

Difference from copyright: Copyright arises automatically upon creation, while design rights require registration. The protection period is 25 years from the design registration application date. Registered designs can be searched using Japan’s J-PlatPat database [10][11].

When the risk materializes: Design rights prohibit third parties from commercially implementing the same or similar design in a product, building, or image. Simply viewing an AI-generated image is not infringement; the risk arises when that design is commercialized or used for business purposes [10].

Design-Right-Specific Considerations

Broad scope of similarity: Rights extend to designs “similar” to the registered design, meaning even slight visual differences may still constitute infringement [10].
Partial designs: A design covering only part of a product (e.g., the button shape of a smartphone) can also be registered [10].
Sets of articles: Designs of multi-component products sold as a set are also protectable [10].

Practical Enterprise Responses

Pre-Use Checklist for Generated Images

□ Copyright Check
  - Visually confirm no close resemblance to a specific copyrighted work
    (illustration, photograph, graphic)
  - For commercial use, confirm the model's terms of service permit commercial use of output
  - Avoid output that intentionally mimics the style of famous works

□ Portrait Rights Check
  - Confirm no elements identifying or specifiable as a real person (celebrity or private individual)
  - Confirm no celebrity's name was used in the prompt
  - Apply especially careful review for commercial use

□ Trademark Check
  - Confirm no existing brand logos or distinctive designs are included
  - Confirm no expression closely resembling a well-known character is included
  - Conduct a trademark search appropriate to the use context (advertising, product packaging, etc.)

□ Design Rights Check (when commercializing or using for business)
  - Confirm the generated product or UI design does not closely resemble a registered design
  - Search J-PlatPat for relevant registered designs before proceeding
  - Confirm no competitor's product design was specified in the prompt
  - Consult IP staff or a patent attorney before commercializing

Designing a Prompt Policy

When leveraging generative AI as an enterprise, it is important to establish policies governing prompts used by employees.

Prohibited Example	Reason
Prompts containing a celebrity’s name	Risk of portrait rights and publicity rights infringement
Instructions to generate a specific brand’s logo	Risk of trademark infringement
Instructions to precisely replicate a specific work or artist’s style	Risk of copyright infringement
Sexual or defamatory depictions of real individuals	Defamation and privacy violation
Instructions to reproduce a competitor’s product design for commercialization	Risk of design rights infringement

Model Selection and Terms of Service Review

The terms of service of generative AI services specify copyright ownership of output, permissibility of commercial use, and prohibited content. Key points to verify before use:

Do rights to output belong to the user, or does the provider retain them?
Is commercial use permitted?
Does the provider have a policy of indemnifying or disclaiming third-party copyright infringement (some services offer “copyright indemnification programs”)?
Is information about training data sources and transparency publicly available?

Record Keeping

To demonstrate appropriate conduct if copyright infringement becomes an issue, the following are recommended to keep on record:

Content of the prompts used
Model and service used, and their version
Date and time of generation, and the intended use
Content of checks performed and their results

Summary

AI-generated images are a collection of intertwined rights issues involving copyright, portrait rights, trademark law, and design rights. No technical means of guaranteeing “legally safe” output has yet been established; ultimately, human review and enterprise policy governance are indispensable. Rights analysis for generative AI is updated through Copyright Office guidance, Agency for Cultural Affairs materials, court decisions, and service terms, so practical use requires ongoing review [2][3][4].

Right	Primary Risk	Key Response
Copyright	Similarity derived from training data	Visual inspection, prompt policy, terms verification
Portrait rights	Identification and commercial use of real individuals	Strict review of images containing people
Publicity rights	Commercial use of celebrity faces and names	Prohibit celebrity-name prompts
Trademark rights	Reproduction of brand logos and trade dress	Trademark research and prohibition policy
Design rights	Similar product, building, or UI designs when commercialized	J-PlatPat search and pre-commercialization IP review
Right to reputation	False or defamatory content	Content-purpose review

References

e-Gov Law Search, Copyright Act (Act No. 48 of 1970)
Agency for Cultural Affairs, View on AI and Copyright
U.S. Copyright Office, Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence (2023)
U.S. Copyright Office, Zarya of the Dawn Registration Review Decision (2023)
Carlini et al., Extracting Training Data from Diffusion Models, arXiv:2301.13188 (2023)
Supreme Court of Japan, Case Search, Grand Bench Decision of December 24, 1969, Minshu Vol. 23, No. 12, p. 1625 (Kyoto Prefectural Student Union case), among others
e-Gov Law Search, Criminal Code (Act No. 45 of 1907)
e-Gov Law Search, Trademark Act (Act No. 127 of 1959)
e-Gov Law Search, Unfair Competition Prevention Act (Act No. 47 of 1993)
e-Gov Law Search, Design Act (Act No. 125 of 1959)
Japan Platform for Patent Information, J-PlatPat

Quiz

Design Prompting Basics