Getty v. Stability AI
12 million images allegedly copied for Stable Diffusion.
Getty Images vs. Stability AI
U.S. District Court, District of Delaware; UK High Court of Justice
Copyright infringement: Stability AI copied 12 million Getty images to train Stable Diffusion withou...
Not specified (seeking injunction + damages)
Getty Images v. Stability AI — The AI Image Generation Case
Case Summary
Getty Images sued Stability AI in January 2023, alleging that Stable Diffusion was trained on 12 million copyrighted Getty images without permission or license. The case is filed in both the U.S. and UK, making it one of the first international AI copyright disputes.
Timeline
| Date | Event |
|---|---|
| Jan 2023 | Getty files in UK High Court |
| Feb 2023 | Getty files in U.S. District Court (Delaware) |
| 2023-2024 | Discovery and motions |
| 2025 | Case progresses through pre-trial |
| Mar 2026 | Still active, trial preparation |
Key Legal Issues
The Training Data Question
Getty alleges Stability AI:
- Copied 12 million copyrighted images from Getty's library
- Used LAION-5B dataset which scraped Getty images
- Generated images that sometimes include Getty watermarks (proving direct copying)
- Created a product that directly competes with Getty's licensing business
The Watermark Evidence
A key piece of evidence: Stable Diffusion sometimes generates images containing distorted Getty Images watermarks. This strongly suggests:
- The model was trained on watermarked Getty images
- The training involved direct copying of copyrighted material
- The model "memorized" elements of the training data
Fair Use Defense
Stability AI argues:
- Training is transformative (learning visual concepts, not copying)
- Generated images are new works, not reproductions
- Similar to how human artists learn from existing images
- No single Getty image is reproduced in outputs
Getty's Counter-Arguments
- Scale matters: 12 million images is wholesale copying
- Commercial purpose: Stability AI profits from Getty's investment
- Market harm: Stable Diffusion directly competes with Getty's licensing
- Watermark reproduction proves non-transformative memorization
Significance
Why This Case Matters for Visual Artists
1. First major case specifically about AI image generation
2. Tests whether training image generators on copyrighted art is legal
3. Could establish licensing requirements for visual training data
4. Affects every AI image tool (Midjourney, DALL-E, etc.)
Potential Industry Impact
- If Getty wins: AI image companies may need to license training data
- Could create new revenue streams for photographers and artists
- May force AI companies to use only licensed/public domain images
- Could reshape the stock photography industry
Related Cases
- Individual artists v. Stability AI (class action)
- Midjourney copyright disputes
- DeviantArt/Stability AI lawsuits
Current Status
ACTIVE — Both U.S. and UK proceedings ongoing. Trial preparation phase as of March 2026.