Getty v. Stability (US)

What Getty alleges

This is the US counterpart to Getty’s London case, alleging Stability copied more than 12 million Getty photographs — along with captions and metadata — to train Stable Diffusion, and that outputs reproduced distorted versions of Getty’s watermarks. Getty filed the two cases within weeks of each other in early 2023, betting on parallel litigation in two jurisdictions with different copyright and trademark frameworks.

The complaint layers a DMCA claim over altered or removed copyright-management information on top of copyright infringement, and adds a full family of trademark claims — infringement, false designation of origin, dilution and unfair competition — tied specifically to watermark artifacts surfacing in generated images. That combination gives Getty multiple independent paths to liability even if any single theory falters.

From Delaware to California: two years of venue before the merits

Getty originally filed in Delaware in February 2023 and spent more than two years in jurisdiction and venue disputes there. On August 14, 2025, Getty voluntarily dismissed the Delaware action and refiled the same day in the Northern District of California — Stability’s home turf — adding new allegations that Stability’s outputs dilute the market for Getty’s own licensed images.

That market-dilution theory now runs through both the unfair-competition claim and the trademark dilution count: Getty argues that AI-generated images competing with its licensed catalog erode the value of its trademarks and its business, not just the copyright in individual photographs. Refiling in California also put the case before a court with a deep bench of AI-related precedent, given how many training-data suits already run through the Northern District.

The ruling and its limits: trademark survives, DMCA claim narrows

In April 2026, the court denied Stability’s motion to dismiss the trademark infringement, false designation of origin, dilution and unfair competition claims. It dismissed only the DMCA false-CMI claim, and did so without prejudice — meaning Getty can attempt to replead it with more specific allegations of intent.

Notably, Stability did not contest the underlying copyright infringement claim at this stage, so this ruling says nothing about the merits of that claim or about fair use. It resolves only the secondary trademark-family theories and the DMCA count — leaving the case’s central copyright question for a later stage of litigation.

“Stability’s motion to dismiss the trademark infringement, false designation of origin, trademark dilution and unfair competition claims is denied; the DMCA false copyright-management-information claim is dismissed without prejudice for failure to adequately plead intent.”— Order on Motion to Dismiss, No. 3:25-cv-06891 (N.D. Cal. Apr. 23, 2026)

Where it stands

The case is in case management and discovery in San Francisco, with a further conference set for November 2026 and no trial date reported. Whether Getty repleads the DMCA claim with adequate intent allegations remains an open question heading into that conference.

The untested copyright infringement claim — the core of Getty’s case — has not yet drawn its own dispositive motion, so the fair-use question that dominates other US training-data suits has not been reached here.

Why it matters for training-data buyers

This is the US case that most squarely pairs a large licensor’s copyright claims with trademark claims over watermark artifacts in outputs. The April 2026 ruling shows an AI developer can face Lanham Act and state unfair-competition exposure for output artifacts even before any fair-use question is decided.

For data suppliers, that supports the position that scraped stock-photo libraries carry copyright and brand risk simultaneously. For buyers evaluating a model or dataset, watermark contamination in generated outputs is now a concrete, litigated diligence item — testable at the output level, not just a paperwork question about how a dataset was licensed at collection time.

The two Getty cases together — narrowed on the merits in London, still fully live on trademark and unresolved on copyright in California — illustrate why buyers can’t treat a single favorable ruling in one jurisdiction as clearing a dataset or model globally.