Two Separate Copyright Questions
Litigation by artists against image generation companies raises two distinct questions that are frequently conflated. The first is whether using copyrighted images to train a model is itself infringement, or whether it is fair use.
The second is whether specific outputs reproduce protected expression from particular works, which is an ordinary substantial similarity question of the kind courts have handled for decades.
Style is not protected by copyright
A claim that a model can produce work in an artist style faces a fundamental obstacle: copyright has never protected style. It protects specific expression. That is why these cases focus on the training copies and on particular outputs rather than on style imitation.
The Legal Arguments
The training claim turns on fair use, weighed across the purpose and character of the use, the nature of the works, the amount used and the effect on the market. Developers argue training is transformative because the model learns statistical relationships rather than storing images.
Artists argue the market effect is severe, since generated output substitutes directly for commissioned work, and that copies were made at scale during the training process regardless of what the model retains.
Output claims are more conventional. Where a generated image reproduces recognisable protected elements of a specific work, ordinary infringement analysis applies without needing to resolve the training question at all.
What Artists Can Do Now
Register works with the Copyright Office. Registration is a precondition to suing in the United States, and timely registration unlocks statutory damages and attorney fees that are otherwise unavailable.
Use available opt-out mechanisms where a dataset or platform offers them, and check licensing terms of any platform you upload to, since some grant broad rights that permit training use.
Check what rights your host platform takes
Portfolio and social platforms frequently take a broad sublicensable licence in uploaded work. That licence, rather than any scraping, may be what permits training use, and it is in terms most artists accepted without reading.
Before You Act
Thinking About Filing a Claim?
Most plaintiff lawyers offer a free initial consultation and work on contingency, meaning no fee unless there is a recovery. LawsuitWatch is not a law firm: we publish explainers, and we do not provide legal advice, representation or referrals. Your state bar directory is the reliable place to find and verify a lawyer.
AI Image Generator Lawsuits: Training Data, Output Similarity and Style: Frequently Asked Questions
Answers to the most common questions about this case and your legal options.
What do AI image lawsuits allege?
That copyrighted images were used to train models without licence, and separately that specific outputs reproduce protected expression from particular works.
Is artistic style protected?
No. Copyright protects specific expression rather than style, which is why claims focus on training copies and particular outputs instead.
What is the fair use argument?
That training is transformative because a model learns statistical relationships rather than storing images, weighed against the market effect on artists.
Are output claims different?
Yes. Where an output reproduces recognisable protected elements of a specific work, ordinary infringement analysis applies without resolving the training question.
What should artists do now?
Register works, since registration is required to sue and timely registration unlocks statutory damages, and check what licence host platforms take.
Legal Disclaimer
This article is general legal information, not legal advice, and does not create an attorney-client relationship. Case status, eligibility criteria, and any amounts described are as reported at the date shown and may change. Consult a licensed attorney in your jurisdiction about your own situation.