Civil Law and AI Image Generation Copyright Enforcement in Europe

1. Introduction

AI image-generation copyright enforcement in Europe concerns disputes arising when generative-AI systems are used to create, reproduce, transform, train on, distribute or commercially exploit images.

Typical disputes include:

an AI model trained on copyrighted photographs without authorisation;

an AI-generated image reproducing a substantial part of an existing artwork;

an output that is extremely similar to a protected photograph or illustration;

a photographer seeking an injunction against use of an AI-generated image;

uncertainty about whether the AI-generated image itself qualifies for copyright;

disputes over whether a detailed human prompt is sufficient to establish authorship;

copying of AI-generated images by another website;

use of copyrighted images as prompts, references or input material;

AI-generated images incorporating protected characters, designs or artistic elements;

disputes over training-data transparency;

enforcement against AI providers that refuse to disclose relevant information.

There is not yet a single EU-wide judicial doctrine specifically governing every AI-generated image. The legal framework is being constructed from existing EU copyright principles, the DSM Copyright Directive, the AI Act, national copyright law and emerging national AI cases. EUIPO itself notes that litigation concerning GenAI copyright remains a developing area. (EUIPO)

The central legal questions are:

1. Is the AI-generated image itself a copyright work?

2. Who, if anyone, is its author?

3. Was a pre-existing copyrighted image used unlawfully during training or generation?

4. Does the output reproduce protected expression from an earlier work?

5. What remedies are available to the copyright holder?

2. The Four Separate Copyright Problems

AI image litigation should not treat everything as one issue.

There are at least four separate legal stages:

Stage 1 — Training/Input

Copyrighted images may be copied or extracted for:

training;

testing;

fine-tuning;

retrieval;

model development.

Stage 2 — Prompting/Generation

A user supplies:

"Create an image of..."

The AI generates an output.

Stage 3 — Output

The resulting image may:

be completely different;

resemble an existing work;

reproduce protected expression;

contain recognisable elements from protected works.

Stage 4 — Enforcement

The output is:

published;

sold;

licensed;

uploaded;

used commercially.

Each stage raises different legal questions.

3. Main European Legal Framework

A. Copyright Directive 2001/29/EC

Directive 2001/29/EC establishes important exclusive rights including:

reproduction;

communication to the public;

making available to the public;

distribution.

These rights are highly relevant when AI systems copy protected images during training or when AI-generated images are subsequently distributed.

4. DSM Copyright Directive 2019/790

The Digital Single Market Copyright Directive is particularly important for AI training.

Article 4 creates a text-and-data-mining exception for lawfully accessible works, but the exception is conditional.

For commercial or other qualifying TDM, rightsholders can expressly reserve their rights, including through appropriate machine-readable means for publicly available online content. (EUR-Lex)

Therefore:

Publicly accessible image ≠ automatically free for unrestricted AI training.

The legality of training depends on:

how the work was accessed;

the purpose of the mining;

whether an exception applies;

whether the rightsholder reserved rights;

national implementation;

other applicable copyright rules.

5. AI Act and Copyright

The EU AI Act adds a specific AI-provider obligation.

For general-purpose AI models, Article 53 requires providers to:

maintain a policy to comply with EU copyright law;

identify and respect rights reservations under Article 4(3) of Directive 2019/790;

prepare and make publicly available a sufficiently detailed summary of content used for model training. (EUR-Lex)

The AI Act specifically recognises that generative AI models may require enormous quantities of:

text;

images;

videos;

other copyright-protected material.

Its Recital 105 explains the connection between AI training, copyright-protected content and TDM exceptions. (AI Act Service Desk)

Important distinction

The AI Act's copyright-compliance requirements do not themselves create a universal rule that every AI-generated image is copyrighted or every training use is unlawful.

Those questions remain governed substantially by copyright law.

6. Copyrightability of an AI-Generated Image

This is one of the most difficult issues.

European copyright law generally protects an original work reflecting the author's own intellectual creation.

The CJEU's case law repeatedly connects originality with the author's own creative choices.

This creates a fundamental distinction:

Human-created image

Human makes the creative choices → photographs/draws/edits → copyright potentially arises.

AI-assisted image

Human makes substantial creative choices → AI assists → copyright may potentially protect the human creative contribution, depending on national law and facts.

Fully autonomous AI output

AI determines the expressive result with insufficient human creative contribution → copyright protection becomes much more difficult.

The EUIPO has identified this question as one of the central unresolved issues surrounding generative AI. (EUIPO)

7. Case 1 — Municipal Court in Prague: S. Š. v Taubel Legal

Case No. 10 C 13/2023-16

Municipal Court in Prague, 11 October 2023

This is one of the most directly relevant European cases.

The claimant used a generative AI system to create an image depicting two people signing a business contract, based on a textual prompt.

The claimant alleged that the defendant law firm subsequently used the image on its website without permission.

The claimant sought:

recognition of authorship;

removal of the image;

an injunction against further infringement.

The court dismissed the claims.

The decision is significant because the court considered whether an AI-generated graphic could constitute a copyright work and whether the person providing the prompt could be recognised as its author. EUIPO identifies the case as the first European ruling concerning copyrightability of AI-generated content. (EUIPO)

Importance

The case demonstrates that:

A person who merely supplies a prompt cannot automatically assume that the resulting AI image belongs to them as a copyright work.

The precise level of human creative contribution matters.

Important qualification

The case was also affected by an evidentiary problem: the claimant did not sufficiently establish that the particular image was generated from the alleged prompt.

Therefore, it should not be overstated as establishing a universal EU rule that all AI-assisted images are uncopyrightable.

It is a national Czech decision, not a CJEU judgment.

8. Case 2 — District Court of Munich, Case 142 C 978625

13 February 2026

This is another particularly important European AI-image case.

The Munich District Court addressed copyrightability of AI-generated images.

According to the EUIPO's case-law summary, the court considered an AI-generated image copyright-protected only where the creative elements incorporated into the prompt dictated the output sufficiently strongly that the resulting work could be regarded as the author's own original creation reflecting free and creative choices. (EUIPO)

Importance

This case is important because it moves the analysis away from the simplistic question:

"Was AI used?"

toward:

"How much creative control did the human exercise over the final expressive result?"

A detailed and creatively structured human contribution may therefore matter more than merely typing a short instruction.

Example

Prompt:

"A dog in a park."

This provides relatively little evidence of human control over the final expression.

Compare:

detailed composition + lighting + perspective + character design + iterative selection + editing + colour decisions + arrangement.

The second situation provides more evidence of human creative choices.

9. Case 3 — Infopaq International A/S v Danske Dagblades Forening

C-5/08

CJEU, 16 July 2009

Infopaq is foundational EU copyright jurisprudence.

The CJEU held that copyright protection is connected to an author's own intellectual creation and examined whether even relatively short portions of protected works could fall within the reproduction right. (Infocuria)

AI relevance

The case is important for AI training because AI systems can process enormous numbers of protected works.

The fact that an AI system copies only parts of an image does not automatically eliminate copyright concerns.

The relevant questions include:

Is the copied material protected?

Is the reproduced part expressive?

Does an exception apply?

Was the copying temporary or permanent?

Was it necessary for lawful TDM?

Principle

Small-scale copying is not automatically outside copyright protection.

This is particularly relevant to AI training processes involving repeated copies of protected images.

10. Case 4 — Eva-Maria Painer v Standard Verlags

C-145/10

CJEU, 1 December 2011

Painer concerned portrait photography.

The CJEU explained that a photograph can constitute an original work where the photographer makes free and creative choices, including choices concerning:

composition;

pose;

lighting;

background;

framing.

The Court recognised that even realistic photographic images can embody creative choices.

AI relevance

This is highly relevant to AI-generated image litigation.

Suppose an AI generates a photorealistic image.

The claimant cannot simply argue:

"It looks realistic, therefore it is not protected."

Painer demonstrates that realism does not prevent copyright protection.

The real issue is:

Where are the human creative choices?

11. Case 5 — Levola Hengelo v Smilde Foods

C-310/17

CJEU, 13 November 2018

Levola established two important principles for identifying a copyright "work":

the subject matter must be original in the sense of being the author's own intellectual creation; and

the protected expression must be identifiable with sufficient precision and objectivity. (Infocuria)

AI relevance

AI-generated images are generally capable of being visually identified with considerable precision.

Therefore, the major difficulty is often not:

"Can the image be identified?"

but:

"Whose original intellectual creation is it?"

That makes the human-authorship question particularly important.

12. Case 6 — Cofemel v G-Star Raw

C-683/17

CJEU, 12 September 2019

Cofemel concerned clothing designs.

The CJEU held that copyright protection cannot depend merely upon aesthetic value. A design must qualify as an original work reflecting the author's own intellectual creation. (curia)

AI relevance

An AI-generated image does not become copyright-protected merely because:

it is beautiful;

it is aesthetically impressive;

it took substantial computing resources to generate;

it has commercial value.

The legally important question remains original creative expression attributable to an author.

13. Case 7 — Brompton Bicycle v Chedech/Get2Get

C-833/18

CJEU, 11 June 2020

Brompton developed the originality test further.

The Court explained that an original work reflects the personality of its author through free and creative choices. Where technical constraints determine the expression, copyright protection may not apply to those technically dictated elements. (Infocuria)

AI relevance

This provides a useful analytical framework for distinguishing:

human creative control

from

technical/system-generated characteristics.

If an AI image's relevant characteristics are almost entirely determined by:

model architecture;

default settings;

random generation;

technical limitations;

it becomes harder to attribute those characteristics to human creative choices.

14. Case 8 — Land Nordrhein-Westfalen v Renckhoff

C-161/17

CJEU, 7 August 2018

Renckhoff concerned a photograph that had originally been lawfully placed online with the photographer's consent.

A student copied the photograph and uploaded it to another website.

The CJEU held that posting the photograph on another website could constitute a new communication to the public requiring authorisation. (curia)

AI relevance

This is highly relevant to AI-generated images once they are commercially distributed.

Example:

Photographer's image → AI training/generation → AI output → website.

Even if the original photograph was publicly accessible, later online use may require separate analysis of the relevant exclusive rights.

Principle

"It was already online" does not automatically mean "any subsequent online use is authorised."

15. Case 9 — Pelham v Hütter

C-476/17

CJEU, 29 July 2019

Pelham concerned musical sampling.

The Court held that unauthorised sampling can infringe the rights of a phonogram producer, while certain uses of a modified sample that is unrecognisable to the ear may fall outside that particular reproduction right. (Infocuria)

AI-image analogy

This provides an important conceptual comparison:

Copying vs transformation

An AI output might transform an input image substantially.

But the fact that an output is "different" does not automatically answer whether protected expression has been unlawfully reproduced.

The court may need to examine:

what was taken;

what was retained;

whether protected expression remains;

whether an exception applies.

This is an analogy, not a case about AI images.

16. Case 10 — Deckmyn v Vandersteen

C-201/13

CJEU, 3 September 2014

Deckmyn concerned the copyright exception for parody.

The CJEU established that parody is an autonomous concept of EU law and identified important characteristics for the exception. (Infocuria)

AI relevance

AI image generators are frequently used for:

parody;

satire;

memes;

transformative artistic works.

Therefore, an AI-generated image that resembles an existing copyrighted image must not automatically be treated as infringement.

Potential exceptions must also be examined.

17. Case 11 — Coty Germany v Stadtsparkasse Magdeburg

C-580/13

CJEU, 16 July 2015

Coty Germany concerned enforcement of intellectual-property rights and access to information concerning alleged infringement.

The CJEU examined the right to information under Article 8 of the Enforcement Directive 2004/48/EC. (Infocuria)

AI relevance

AI copyright litigation often involves an information problem.

A rightsholder may know:

"My images appear to have been used."

But may not know:

whether the AI model trained on them;

when they were copied;

how they were processed;

which model was used;

whether the output derives from them;

who distributed the resulting image.

Enforcement mechanisms concerning access to relevant information can therefore become particularly important.

18. AI Training and Copyright

The training stage deserves separate analysis.

Suppose an AI provider downloads:

100 million online images.

It creates a dataset.

The images are:

copied;

processed;

analysed;

stored;

potentially discarded;

incorporated into training processes.

The legal analysis should ask:

Question 1

Were the images lawfully accessible?

Question 2

Does Article 3 or Article 4 DSM Directive apply?

Question 3

Was the activity commercial or scientific?

Question 4

Did the copyright owner reserve TDM rights?

Question 5

Was the reservation machine-readable where required?

Question 6

Was another licence obtained?

Question 7

Did the AI provider comply with its AI Act obligations?

Article 4 expressly permits certain reproductions/extractions for TDM but conditions the broader exception on the absence of an appropriate rights reservation. (EUR-Lex)

19. Training Is Not the Same as Output Infringement

This distinction is extremely important.

Situation A — Lawful training, infringing output

Training might fall within a lawful exception, but an output could independently reproduce a protected image.

Situation B — Unlawful training, non-infringing output

Training may raise copyright issues even though the final output is sufficiently different.

Situation C — Lawful training and non-infringing output

There may be no copyright infringement.

Situation D — Unlawful training + substantially reproducing output

Two separate copyright problems may potentially arise.

Therefore:

Training legality and output legality must be analysed separately.

20. AI Output That Looks "Similar"

Similarity alone does not automatically establish copyright infringement.

Copyright generally protects expression, not abstract ideas or styles.

For example:

"Create a fantasy castle in a sunset."

Many different images could result.

A claimant would generally need to identify protected expression rather than merely saying:

"The AI copied my artistic style."

21. Style Imitation

One of the most controversial questions is:

Can an artist prevent an AI from generating images "in their style"?

EU copyright law does not generally create a broad exclusive copyright over an abstract artistic style.

The analysis may instead concern:

actual copying;

protected characters;

specific compositions;

recognisable expressive elements;

photographs;

trademarks;

personality rights;

unfair competition;

national laws concerning passing off or personality rights.

Therefore:

"Same style" and "same protected expression" are not automatically equivalent.

22. Substantial Similarity

Suppose an AI produces an image that is almost identical to a famous illustration.

The court may ask:

Is the original illustration protected?

What elements are original?

Are those elements reproduced?

Was the reproduction authorised?

Does an exception apply?

Is the AI output sufficiently connected to the protected work?

What rights were infringed?

The comparison should focus on protected expressive elements, not merely general themes.

23. Human Prompt as Copyright

A prompt itself can potentially raise a separate question.

For example:

"A cinematic portrait of an elderly astronaut standing on a deserted lunar station, dramatic side lighting, deep shadows, reflective helmet, surreal architectural composition..."

Is that prompt itself a copyright work?

Potentially, if it reaches the necessary originality threshold under applicable national law.

But:

A copyright-protected prompt does not automatically give the author copyright over every AI output generated from it.

The image and the prompt are legally distinct subject matter.

24. Human Editing After AI Generation

This is one of the most important practical areas.

Suppose:

AI generates basic image

↓

Human:

changes composition;

paints over portions;

changes lighting;

adds characters;

removes elements;

combines several outputs;

performs substantial digital editing.

The final work may contain:

AI-generated material + human-authored material.

Copyright analysis may therefore focus on the human contribution.

This is consistent with the broader EU originality doctrine: protection depends upon identifiable creative choices attributable to the author.

The EUIPO's GenAI study specifically identifies the extent of human involvement as an important issue. (EUIPO)

25. AI Output and Derivative Works

A particularly difficult situation is:

Original copyrighted image → AI transformation → new image.

Examples:

changing clothing;

changing background;

changing colour;

converting photograph into painting;

altering facial expression;

adding objects.

The legal question is whether the resulting image unlawfully reproduces or transforms protected expression.

The fact that the output is technically "new" does not automatically eliminate the copyright issue.

26. Enforcement Remedies

Under European copyright enforcement rules, available remedies can include:

1. Injunction

Order requiring the defendant to stop:

reproducing;

distributing;

communicating;

selling;

displaying the infringing image.

2. Removal

The infringing image may be removed from:

websites;

marketplaces;

advertising platforms;

social-media accounts.

3. Damages

Depending upon national implementation and circumstances, the rightsholder may claim compensation.

4. Information

The rightsholder may seek information concerning:

source;

distribution;

commercial channels;

infringing activity.

5. Evidence preservation

Digital evidence can be crucial.

6. Destruction/withdrawal

National implementation may permit measures concerning infringing copies and materials.

27. Proof in AI Copyright Litigation

AI disputes create unusual evidentiary challenges.

A claimant should preserve:

Original image

high-resolution file;

metadata;

RAW file where available;

creation date.

AI evidence

prompt;

model;

model version;

generation date;

seed;

settings;

iterations;

image-to-image inputs.

Training evidence

Where available:

dataset information;

crawling records;

licensing information;

model documentation;

transparency disclosures.

Publication evidence

screenshots;

URLs;

dates;

advertisements;

sales records.

28. AI Act Training-Data Transparency

Article 53 requires general-purpose AI providers to publish a sufficiently detailed summary of training content.

This is important for enforcement because it may give rightsholders a better basis to investigate whether their protected works were used.

The AI Act specifically states that this transparency is intended to help copyright holders exercise and enforce their rights. (EUR-Lex)

However:

A training-data summary is not necessarily a complete itemised list of every image used.

The statutory requirement concerns a sufficiently detailed summary, while current policy discussions continue concerning deeper transparency.

The European Parliament in March 2026 called for stronger transparency and discussed more detailed mechanisms, but those recommendations should not be confused with binding amendments already in force. (EUR-Lex)

29. Deepfakes and Copyright

AI image generation may also involve deepfakes.

The AI Act contains transparency requirements for certain AI-generated or manipulated image, audio and video content constituting deepfakes.

The relevant obligations concern disclosure that content was artificially generated or manipulated, subject to the conditions and exceptions in the Act. (EUR-Lex)

This is not itself a copyright rule.

An image can therefore involve:

AI Act issue + copyright issue + personality/privacy issue

simultaneously.

30. Personality Rights and Images

Copyright is not the only relevant legal right.

AI-generated images may depict:

celebrities;

private individuals;

politicians;

employees;

children.

Potential additional claims can involve:

image/personality rights;

privacy;

data protection;

defamation;

trademark;

unfair competition.

These rights vary considerably between European jurisdictions.

Therefore, an image that is not copyright-infringing can still potentially create other civil-law problems.

31. AI Provider vs AI User

Liability may involve different actors.

The existence of a contract with an AI provider does not necessarily guarantee that the resulting image is free from third-party copyright claims.

32. AI Provider Terms and Conditions

The user should distinguish:

Contractual permission

The AI provider may tell the customer:

"You may commercially use your output."

from:

Third-party copyright

The provider may not necessarily own or control all rights potentially implicated by the output.

Therefore:

A platform's terms of service cannot automatically extinguish a third party's copyright.

A commercial user should independently consider infringement risk.

33. Copyright Infringement Formula

A useful analytical formula is:

Protected Original Work + Unauthorised Relevant Act + Protected Expression Reproduced/Communicated + No Applicable Exception = Potential Copyright Infringement

For AI:

Training Copying + Protected Work + No Applicable TDM/Licence/Exception = Potential Training-Stage Claim

For output:

AI Output + Reproduction of Protected Expression + Unauthorised Commercial Use = Potential Output-Stage Claim

34. Important Distinction: Copyrightability vs Infringement

These are different questions.

Copyrightability

Is the AI-generated image itself protected?

Infringement

Does the AI-generated image unlawfully reproduce somebody else's protected work?

An image can theoretically be:

uncopyrightable itself but still unlawfully reproduce someone else's protected work, depending on the relevant rights and circumstances; or

copyrightable because of human creative contribution while also infringing an earlier work.

Therefore:

Ownership and infringement must be analysed separately.

35. Important Distinction: AI Image vs AI Model

Another critical distinction:

AI model

The underlying technology.

Training dataset

The collection of images/data used during development.

Prompt

Human instruction.

Output

Generated image.

Edited output

AI output modified by a human.

Each can have a different legal status.

36. Enforcement Scenario

Suppose a photographer discovers that an AI provider used 50 of her photographs in training.

Step 1

Establish ownership of the photographs.

Step 2

Establish that the photographs are protected works.

Step 3

Investigate whether they were lawfully accessible.

Step 4

Determine whether TDM Article 3/4 applies.

Step 5

Check whether rights were reserved.

Step 6

Examine the AI provider's copyright policy and AI Act compliance.

Step 7

Determine whether actual copies were made.

Step 8

Examine whether outputs reproduce protected expression.

Step 9

Preserve evidence.

Step 10

Seek appropriate civil remedies.

37. Important European Cases — Consolidated Table

38. The Two Direct AI Cases

For an examination or research paper specifically about AI image-generation copyright, the two particularly important European national cases are:

1. Prague Municipal Court — 10 C 13/2023-16

AI-generated image + prompt + authorship + infringement claim.

2. Munich District Court — 142 C 978625

AI-generated image + human creative choices + originality.

These should be clearly identified as national decisions, not CJEU authorities. The broader EU legal framework is supplied by cases such as Infopaq, Painer, Levola, Cofemel and Brompton Bicycle. (EUIPO)

39. Current Legal Position in Europe

As of September 2026, the safest way to state the position is:

A. Pure AI output

Copyright protection is uncertain and often difficult to establish where there is insufficient human creative contribution.

B. AI-assisted human creation

Potentially stronger case for copyright where the human exercises substantial and identifiable creative choices.

C. AI training

Not automatically lawful merely because images are publicly accessible.

The DSM Directive's TDM provisions and rights-reservation mechanism are central. (EUR-Lex)

D. AI provider obligations

General-purpose AI providers have specific copyright-policy and training-content-summary obligations under Article 53 AI Act. (EUR-Lex)

E. Output infringement

Still requires examination of the protected work, the allegedly copied expression, the relevant exclusive right and any applicable exception.

F. Enforcement

Traditional copyright remedies remain important, but AI creates additional evidence and attribution difficulties.

40. Practical Civil-Law Analysis

A claimant should ask:

1. Ownership

Who owns the original image?

2. Originality

What makes the image an original work?

3. AI involvement

Was AI used merely as a tool or did it independently determine the expressive result?

4. Human contribution

What creative choices did the human make?

5. Training

Was the original work used during training?

6. TDM exception

Did Article 3 or 4 DSM Directive apply?

7. Rights reservation

Did the rightsholder opt out?

8. Output similarity

Does the output reproduce protected expression?

9. Commercial use

How was the output exploited?

10. Remedies

What injunction, damages, information or removal remedy is available under the applicable national implementation?

41. Key Legal Problems for Future Litigation

Future European litigation is likely to focus heavily on:

whether prompts constitute creative authorship;

how much human control is sufficient;

whether training copies are covered by TDM exceptions;

machine-readable opt-outs;

evidence of training-data use;

memorisation of particular images;

output similarity;

AI-generated adaptations;

copyright in AI-assisted works;

copyright infringement through image-to-image systems;

dataset licensing;

compensation;

cross-border jurisdiction;

disclosure of training information.

EUIPO's 2025 GenAI study expressly identifies both the input/training phase and the generation/output phase as separate copyright questions. (EUIPO)

42. Conclusion

AI image-generation copyright enforcement in Europe is currently built on traditional copyright principles applied to a new technological process.

The most important principles are:

AI use does not automatically create copyright in the resulting image.

Human creative choices remain central to EU originality doctrine.

A detailed prompt alone does not automatically establish authorship.

AI training and AI output are separate copyright questions.

Public availability of an image does not automatically mean unrestricted AI-training permission.

Article 4 DSM Directive provides a TDM exception subject to important conditions, including rights reservation.

Article 53 AI Act imposes copyright-policy and training-summary obligations on general-purpose AI providers.

An AI output can create infringement issues even when it is described as "new" or "generated."

Style imitation and copying protected expression are not automatically the same thing.

Human editing and creative selection can materially change the copyright analysis.

Copyright enforcement requires strong evidence concerning authorship, prompts, datasets, model outputs and publication history.

National courts are beginning to address AI-generated-image copyrightability directly, but there is not yet a comprehensive CJEU ruling specifically resolving AI-image authorship.

Core Formula

Protected Work + AI Use + Relevant Reproduction/Communication + No Licence or Applicable Exception + Protected Expression + Evidence = Potential Copyright Claim

For AI-generated output:

Human Creative Contribution + Original Expression + Identifiable Authorship = Potential Copyright Protection

For AI training:

Lawfully Accessible Work + TDM Activity + Rights Reservation/Exception Analysis + Actual Copying = Training-Stage Copyright Analysis

Ultra-Short Revision

Prague 10 C 13/2023-16 → AI image + authorship
Munich 142 C 978625 → human creative choices + AI image
Infopaq C-5/08 → originality/reproduction
Painer C-145/10 → creative choices in photographs
Levola C-310/17 → identifiable work + originality
Cofemel C-683/17 → originality, not mere aesthetic value
Brompton C-833/18 → free and creative choices
Renckhoff C-161/17 → online image reposting
Pelham C-476/17 → reproduction/transformation
Deckmyn C-201/13 → parody exception
Coty Germany C-580/13 → enforcement/information

Main principle: European AI-image copyright disputes turn less on the mere fact that AI was used and more on the identifiable human creative contribution, the legality of the input/training process, the protected expression contained in the output, and the evidence available to enforce the relevant rights.