What Happened
On January 17, 2023, Getty Images filed a lawsuit against Stability AI in the High Court of London (and a related action in the U.S.) alleging that Stability AI copied and processed over 12 million Getty Images photographs to train its Stable Diffusion image generation model without obtaining a license. Getty presented evidence that Stable Diffusion could generate images containing distorted versions of the Getty Images watermark, strongly suggesting the model had ingested watermarked training images.
The lawsuit claimed copyright infringement, violation of the DMCA (by removing watermark metadata), and trademark infringement. Getty argued that Stability AI built a commercial product that directly competes with Getty’s licensing business using Getty’s own copyrighted content as the foundation.
Why It Matters
This case represents the visual arts equivalent of the NYT v. OpenAI text copyright battle. It tests whether training an AI model on copyrighted images constitutes infringement and whether the resulting model’s ability to generate competing images constitutes derivative works. The distorted Getty watermarks appearing in outputs provided unusually direct evidence of training data provenance. The outcome has implications for every AI image generation company and for the business models of stock photography, illustration, and visual arts broadly.
Lessons Learned
AI companies cannot assume that scraping publicly accessible images is legally permissible for commercial training. Watermark artifacts in model outputs provide forensic evidence of training data composition. The competitive displacement argument is stronger when AI outputs directly substitute for the copyrighted training inputs. Companies building generative AI should secure licenses proactively rather than face litigation retroactively.
Current Status
Getty Images and Stability AI reached a settlement in late 2024 with undisclosed terms. Stability AI subsequently entered licensing agreements with multiple image providers. The company also faced a separate class action from individual artists. The case influenced the broader AI industry toward licensing agreements with content creators, with companies like Adobe, Shutterstock, and others building AI tools trained exclusively on licensed or consent-based datasets.