How to Protect Your Art From Being Used to Train AI

AI image generators were trained on billions of images scraped from the web — and unless you posted your work after 2023, yours is almost certainly in one of those datasets. Here's what you can actually do about it.

How AI Training Scraping Works

Companies building image generators crawled the public web and downloaded everything they could find. robots.txt — the file that tells crawlers to stay out — was largely ignored, and courts haven't ruled definitively that ignoring it is illegal.

Once your art is in one of these datasets, it trains the model to generate images in your style, on demand, for anyone. No credit, no payment, no permission asked.

1. Embed Copyright Metadata in Every File

Every image file supports EXIF and XMP metadata fields for copyright, creator, and usage terms. Embedding "Not for AI training" in these fields is a clear legal notice — even if crawlers don't read it today, it strengthens any future legal claim. Vulta automatically injects this metadata into every file it processes.

2. Use Invisible Forensic Watermarking

A forensic watermark hides a unique ID inside your image at the pixel frequency level — invisible to the eye, but detectable by scanners even after the image has been compressed or resized. If your work ends up in a training dataset, the watermark gives you evidence pointing directly to which file was scraped and who scraped it. That's what Vulta does when you upload.

3. Only Share Low-Resolution Previews

Share a resized version (1000–1200px on the long edge) on social media and portfolio sites rather than full resolution. Models trained on small images produce worse results, and low-res files can't be used for print-on-demand products — two reasons scrapers prefer to skip them.

4. Register on Opt-Out Lists

Some AI companies let you opt out of future training runs via Spawning's "Have I Been Trained?" database. It won't remove you from datasets that already exist, but it stops participating companies from including you in the next round. Worth five minutes of your time.

5. Anchor Your Work Before You Publish

Anchoring your work on the blockchain before you post it creates a timestamped record that predates any training dataset. If legal action around AI training ever becomes viable — and many IP lawyers think it will — that timestamp is going to matter.

How Vulta Handles This

Vulta stacks all of these protections together: invisible watermarks for detection, copyright metadata baked into every file, blockchain timestamps for prior-creation proof, and a no-AI-training clause built into every delivery contract. Here's how it all works.