Stability AI v. Getty Images February 2023: When Image Generators Faced Their First Major Training-Data Copyright Lawsuit

Incident summary

On 16 January 2023 Getty Images filed suit against Stability AI in the High Court of Justice of England and Wales. On 3 February 2023 Getty filed a parallel suit in the United States District Court for the District of Delaware (Case No. 1:23-cv-00135). Both complaints allege that Stability AI used Getty's copyrighted image corpus, including watermarked images, in training the Stable Diffusion text-to-image model without license or authorization.

The US complaint alleged copyright infringement, false designation of origin, trademark dilution, and unfair competition. Getty submitted evidence including Stable Diffusion outputs containing distorted versions of the Getty Images watermark, presented as evidence that the watermarked images had been included in training and that the model had partially memorized the watermark pattern.

Stability AI's public response denied infringement and asserted defenses including fair use (US) and text-and-data-mining exemptions (UK and EU). Both cases proceeded through 2023-2024 with motions and discovery; trial and final resolution timelines extended into 2025-2026.

Failure technique

The legal-technical pattern is reuse of copyrighted training material without licensing in a generative-AI context where the model can produce outputs that resemble specific training examples. Stable Diffusion's training corpus (LAION-5B) included Getty-watermarked images that the corpus assembly process did not filter out.

The Getty watermark evidence is the most novel aspect of the case. By demonstrating that Stable Diffusion outputs contained distorted Getty watermarks, Getty established that watermarked images had been in the training corpus and that the model had partially memorized the watermark. The argument bridges from corpus composition to model output, making the training-data question concrete in a way the parallel Authors Guild v. OpenAI text-generation case cannot easily replicate.

Per the US Copyright Office July 2024 report on Generative AI Training, the legal questions are open and the litigation is foundational. The Office report identifies the Getty v. Stability case as one of the cases whose resolution will shape US copyright doctrine for AI training data.

Impact and consequences

Direct commercial impact on Stability AI during 2023-2024 was substantial: the company faced parallel jurisdictional litigation in two major IP forums simultaneously while also handling leadership transitions and competitive pressure from later image generators. The legal cost and operational distraction shaped the company's trajectory.

Industry impact: the Getty case is the foundational training-data copyright case in the image-generation domain. Subsequent commercial image-generator launches (Adobe Firefly with its commercially-safe positioning, OpenAI DALL-E 3 with explicit safety positioning, Google ImageFX) have all addressed training-data provenance and indemnification more explicitly than 2022-era products did. Adobe Firefly's positioning is largely a market response to the Getty v. Stability litigation.

Legal-precedent impact: trial and final resolution of the Getty cases is one of the foundational events shaping US and UK AI training-data copyright doctrine in 2025-2026. The Authors Guild v. OpenAI and NYT v. OpenAI text-generation parallels will resolve similar questions for text training data.

Lessons for builders

Treat training-data IP review as a launch-readiness gate for commercial image-generator products. The post-Getty market shifted to explicit licensing (Adobe Firefly), explicit indemnification (Adobe, OpenAI), or open-licensing-only training data (some open-weight model providers). AI Strategy Lead and Senior AI Product Manager own this gate.

Document training-corpus composition with sufficient specificity that licensing claims can be defended. Adobe Firefly's specific claim that Firefly was trained on Adobe Stock content (later qualified per the Bloomberg April 2024 reporting) is the operational model.

Watch the model-output side for memorization evidence. Getty's evidence of distorted watermarks in Stable Diffusion outputs is what bridged from corpus claim to model claim. Pre-deployment evaluation should test for memorization of identifiable training-corpus signatures (watermarks, specific photographer styles, recognizable image regions).

Track US Copyright Office and EU AI Act developments. The legal environment is actively shaping; foundation-model providers operating commercially need a maintained understanding of the regulatory and case-law surface.

Mitigations

What builders should put in place to address the failure pattern. Each mitigation maps to operational practice the relevant Applied AI roles own.

Related Applied AI roles

The Applied AI roles whose day-to-day work would have prevented, detected, or contained this incident.

• AI Strategy Lead: An AI Strategy Lead owns organizational AI strategy and prioritization at the company level.
• Senior AI Product Manager: A Senior AI Product Manager owns AI product strategy across multiple feature areas.
• AI Product Manager: An AI Product Manager owns AI-powered product features and the roadmap that ships them.
• AI Engineer: An AI Engineer builds production cybersecurity-relevant AI systems integrating LLMs, embeddings, and retrieval pipelines.

Related AI Decipher Files

• New York Times v. OpenAI (Dec 2023): The Copyright Case That Defines AI Training Liability
• Stanford Internet Observatory LAION-5B CSAM Discovery December 2023: When an Open Pretraining Corpus Failed Provenance Review
• Adobe Firefly Training Data Controversy April 2024: When 'Ethically Trained' Claims Met Disclosed Use of AI-Generated Images

Frequently asked questions