Recommended Free Tools
Crashes, No Sound, or Screen Glitches?
Free scan · under a minute
PC Slower Than It Used to Be?
Free scan · Windows 10 & 11
Outdated Drivers Are Slowing You Down
Free scan · exact hardware match
ThatPainter is reader-supported. When you buy through links on our site, we may earn an affiliate commission. Learn More
Artists did not win a final copyright judgment against AI image generators. The “big win” refers to a discovery-stage ruling in Andersen v. Stability AI Ltd. that allowed important copyright allegations to continue instead of being dismissed before the artists could examine the defendants’ evidence.
The proposed class action concerns Stability AI, Midjourney, DeviantArt and, in later pleadings, Runway AI. The artists may now seek evidence about training data, model development and allegedly infringing outputs. But the court has not ruled that AI training is categorically unlawful, awarded damages, ordered a shutdown, or required models to be destroyed.
Free tools Windows power users keep installed
One-click scans. No signup required.
Which lawsuit is this?
The headline refers to Andersen et al. v. Stability AI Ltd. et al., No. 3:23-cv-00201, in the U.S. District Court for the Northern District of California. Sarah Andersen, Kelly McKernan and Karla Ortiz filed the proposed class action in January 2023.
#1 Best Overall
The original complaint named:
- Stability AI, associated with Stable Diffusion;
- Midjourney, whose service generates images from user prompts;
- DeviantArt, which offered the DreamUp image-generation product.
Runway AI appears in later versions of the litigation. These companies should not be treated as though they all performed the same act: one defendant may be accused of assembling training data, another of developing a model, another of distributing a service, and another of facilitating user-generated outputs. The federal case record and case chronology reflect that the pleadings and defendant lineup developed over time.
What did the artists allege?
The artists alleged that copyrighted artwork was collected from the internet and included in datasets used to develop AI image systems. Their theories included claims that:
- training involved unauthorized copying of protected artworks;
- Stable Diffusion was built substantially from copyrighted images;
- the systems could produce images containing protected expression or images substantially similar to works in the training data;
- users could imitate named artists or commercially compete with them; and
- the defendants benefited from, induced or facilitated infringement by users.
Those are allegations by the plaintiffs, not findings that every image was copied, that all generated images infringe, or that the companies deliberately stole every work mentioned in the complaint. The original complaint is useful for understanding the claims, but it is not proof that those claims will succeed.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →What did the judge actually allow?
After dismissing substantial portions of the original case, Judge William Orrick allowed important copyright-related allegations to proceed into discovery. The surviving theories focused particularly on whether Stability AI’s alleged copying and use of training images in developing Stable Diffusion could constitute infringement, and whether the system’s operation could invoke or reproduce protected material.
The court also considered allegations that the system could facilitate infringement “by design.” That point means the plaintiffs had pleaded a legally plausible theory at that stage. It is not a final finding that the product was designed to infringe.
Why “survived dismissal” matters
A motion to dismiss asks whether a complaint has plausibly stated a legal claim, generally assuming well-pleaded factual allegations are true for purposes of that motion. It does not decide whose evidence is accurate.
Discovery is the next evidence-gathering phase. The parties can request documents, technical information, communications, data and testimony that may show what happened. Only later motions, a trial or a settlement can resolve the underlying liability questions.
Rank #2
What claims were dismissed or weakened?
The artists did not receive permission to pursue the entire original case unchanged. Reported rulings dismissed or narrowed several theories, including:
- claims against defendants whose alleged connection to Stability AI’s training process was not adequately pleaded;
- some vicarious- or induced-infringement theories at earlier stages;
- Digital Millennium Copyright Act claims involving removal or alteration of copyright-management information; and
- unjust-enrichment theories treated as preempted by copyright law, subject to limited opportunities for further pleading.
Because the plaintiffs amended their complaint, the precise list of surviving claims changed over time. The safest summary is that meaningful direct copyright allegations survived, while broad theories against every defendant and several auxiliary claims did not. Courthouse News’ account of the discovery ruling describes the principal procedural development.
Why discovery could decide the case
The dispute now depends less on headlines and more on technical and documentary evidence. Important questions include:
How was the training data collected?
- Which datasets were used?
- Did they contain identifiable works by the named artists?
- Were images downloaded, stored, transformed, captioned or otherwise processed?
- What did the defendants know about copyright status?
- Were licenses, permissions or opt-out systems available?
What does the model retain or reproduce?
- Can the model generate close reproductions of particular training images?
- Does it retain protected expression, or only statistical relationships learned from examples?
- Are near-duplicates rare failures or predictable results?
- Do prompts naming artists produce recognizable features of their work?
- What role do model weights, text encoders, image encoders and the user interface play?
Which defendant did what?
The evidence must also connect conduct to a particular defendant. A company alleged to have created a training dataset may face a different theory from a platform that distributes a model, embeds another company’s technology or gives users an interface for generating images. Proof involving Stable Diffusion’s development would not automatically establish liability for Midjourney, DeviantArt or Runway AI.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Training copies and generated outputs are separate questions
The case involves at least two related but distinct issues.
- Training: Did reproducing copyrighted images while building or training a model infringe copyright, or was the copying protected by fair use or another doctrine?
- Outputs: Does a particular generated image reproduce protected expression from a specific copyrighted work, and who is legally responsible for that result?
Proving one does not automatically prove the other. A plaintiff might establish that a copy was made during training without proving that every later output infringes. Conversely, a close output may raise a separate copying and substantial-similarity dispute without answering every question about the training process.
The major legal issues
Is copying for training fair use?
In the United States, fair use is likely to be a central defense. Section 107 directs courts to consider the purpose and character of the use, the nature of the copyrighted work, the amount and substantiality copied, and the effect on the market for the original or related licensing markets. See the U.S. Copyright Act, 17 U.S.C. § 107.
Defendants are expected to argue that model training is technologically transformative, that the model does not function as a conventional archive of image files, and that the use serves a different purpose. The artists are expected to emphasize commercial exploitation, unauthorized copying and competition with markets for artwork and licensing.
Calling a use “transformative” does not end the analysis. The Supreme Court’s decision in Andy Warhol Foundation v. Goldsmith underscores that courts may examine commercial purpose and market substitution rather than treating a new aesthetic or medium as automatically decisive.
Does visual similarity prove infringement?
No. A similar mood, genre, palette or general artistic style is not by itself proof that protected expression was copied. Copyright generally does not give an artist a monopoly over an artistic style.
A stronger claim would ordinarily focus on a particular protected work, evidence of access or copying, substantial similarity in protected expression, and the defendant’s role in producing or distributing the result. A viral resemblance can be compelling evidence for investigation, but it may not satisfy the legal test on its own. Technical testing and expert analysis may be necessary to determine whether a model produced a near-duplicate or merely a work with shared general characteristics.
Is the AI-generated image itself copyrightable?
That is a separate issue. Courts must distinguish between:
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstall- whether the original artist’s work was protected;
- whether the work was copied during training;
- whether an output reproduced protected expression;
- whether the output contains enough human authorship to receive copyright protection; and
- which party, if any, is responsible for infringement.
An artist can have copyright in a human-created source image even if a later machine-generated image receives little or no copyright protection itself.
Do not overread this ruling
- It is not a final finding of infringement. The artists still must prove their claims with evidence.
- It is not a nationwide ban on AI training. No general rule requiring permission for every training use was established.
- It does not shut down Stable Diffusion or other image generators. No model-destruction or shutdown order was identified.
- It does not make every style imitation illegal. Style, by itself, is not the same as protected expression.
- It does not mean the case is a certified class action. It should be described as a proposed class action unless certification is independently verified.
- It does not automatically decide claims against every AI company. Liability depends on the defendant, product, conduct and evidence.
How the Getty and Hollywood cases fit in
Getty Images in the United Kingdom
Getty’s UK litigation against Stability AI is a separate proceeding. Getty alleged that Stability AI used millions of Getty images to train Stable Diffusion and that some outputs reproduced Getty watermarks.
Getty abandoned its primary copyright allegations during the trial. The court rejected most of the remaining case but found limited trademark infringement involving a Getty watermark. The result was narrow, not a general ruling that all AI training is lawful. The Associated Press reported that Stability AI largely prevailed on the copyright dispute while Getty obtained the limited trademark finding. Read the AP report on the UK result.
Getty also has separate U.S. litigation against Stability AI. The UK trademark finding should not be treated as an answer to the U.S. copyright questions in Andersen.
Recommended Free Tools
Disney, Universal and Warner Bros. versus Midjourney
Major entertainment companies have separately sued Midjourney over alleged generation and distribution of recognizable copyrighted characters. Those cases concern commercial exploitation of famous entertainment properties and are not the same lawsuit as Andersen. A development in one case does not establish that the artists in another case have won.
The broader landscape includes disputes involving books, journalism, software, music and visual media. The categories overlap technologically, but a ruling involving one type of work may not control another. AP reported in 2026 that AI companies faced more than 50 copyright lawsuits, illustrating the scale of the issue without resolving any individual claim.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What happens next?
The expected path is fact-intensive:
- Discovery: The parties exchange documents, technical records, data and testimony.
- Expert analysis: Specialists may examine datasets, model architecture, outputs and reproducibility.
- Summary judgment: Either side may ask the judge to resolve claims without a trial if the evidence shows no genuine dispute of material fact.
- Settlement or narrowing: Some claims may be resolved while others continue.
- Trial: Remaining factual disputes may be presented to a jury or judge.
- Appeal: Important legal questions could move to a higher court.
Secondary case reporting available through August 18, 2026 listed September 8, 2026 as a trial date. Because this article is being prepared before that date and trial settings can change, the date should be checked against the live federal docket immediately before publication. It should not be presented as operative without that confirmation.
What artists and commercial users should take from it
For artists and rights holders
- Preserve dated originals, drafts, metadata, publication records and licensing history.
- Document allegedly similar outputs alongside the specific source works they resemble.
- Keep evidence of prompts, dates, accounts, URLs and reproducibility where lawful and available.
- Separate a claim about a particular work from a broader objection to style imitation.
- Consider registration, ownership records and jurisdiction before assessing litigation options.
For commercial AI users
“Commercial use allowed” is not the same as a guarantee that every output is free of third-party claims or that the output will receive copyright protection. Before adopting a tool, review:
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitches- commercial-use and ownership terms;
- the scope of any indemnity, including plan, jurisdiction and use-case limits;
- training-data representations and provenance disclosures;
- filters for copyrighted characters, logos and artist-name prompts;
- retention, privacy, audit-log and enterprise-administration controls; and
- whether the product is hosted-only, available by API or deployable locally.
For businesses already working in Adobe’s ecosystem, Adobe Firefly may be worth comparing for workflow integration and commercially oriented controls. That is a product consideration, not a legal safe harbor: buyers should still read the current terms and assess their own use case. Midjourney, Stability AI, Recraft and Black Forest Labs’ FLUX may suit different creative or technical workflows, but their licensing, provenance and indemnity terms should be checked directly and separately.
Do not choose a generator solely on image quality or monthly credits. Compare commercial rights, output terms, indemnity, safeguards, provenance, deployment options and the risk created by your intended use—especially recognizable characters, logos, near-duplicates or prompts naming living artists.
Frequently Asked Questions
Did the artists win their copyright case against AI image generators?
No. They won an important procedural victory: key claims survived dismissal and could proceed to discovery. The court has not entered a final judgment finding copyright infringement.
Does the ruling make AI training illegal?
No. It does not establish a general rule that all AI training is unlawful. The ultimate answer may depend on the data, copying process, model, defendant, market effects and evidence in the case.
Can artists sue over an AI image that resembles their style?
Similarity in style alone is generally not enough. A stronger claim would identify a specific protected work and address copying, access, substantial similarity and the defendant’s role.
What should businesses check before using an AI image generator commercially?
Review commercial-use rights, output ownership, indemnity scope, training-data disclosures, safeguards, retention policies, enterprise controls and deployment options. Commercial permission is not a guarantee against every third-party claim.
The Bottom Line
The artists moved Andersen v. Stability AI from a case that might have ended before evidence was collected into serious, fact-based litigation. That is a meaningful win—but it is not proof that the defendants infringed copyright, not a ban on AI image generators, and not a universal answer to whether training models on copyrighted art is lawful.
Quick Recap
