How to Stop AI from Using Your Photos for Training
You post a photo on Instagram. Months later, you learn it may have been used to train an AI image generator — your face, your kids, your holiday snaps, absorbed into a model that can now generate images “in the style of” your life. Creepy? Many people think so. If you’re asking how do I stop AI from using my photos for training, here’s the honest truth: you can’t fully prevent it, but you can significantly reduce your exposure. Here’s what actually works.
The Uncomfortable Reality First
AI image models were trained on billions of photos scraped from the public web — Flickr, stock sites, blogs, social media, anywhere images were accessible. If your photos were publicly visible online, they may already be in training datasets like LAION. You can’t retroactively remove them from models already trained. What’s done is done — but you can control what happens from today forward, and you have more leverage than you might think.
Step 1: Lock Down Your Social Media Privacy
This is the highest-impact move:
- Set accounts to private where possible — private Instagram and Facebook accounts are far harder to scrape at scale than public ones.
- Review what’s already public. Audit your public posts; consider deleting or privatising old public photo albums, especially those featuring children.
- Check platform AI settings. Meta uses public Facebook and Instagram content for AI training — look for the objection/privacy settings in your region (UK users have GDPR-based rights to object). Other platforms have similar settings buried in privacy menus; they change often, so check periodically.
- Disable photo tagging by others or require approval — friends’ public posts of you are a scraping vector you don’t control directly.
Our guide on opting out of AI training on social media walks through the platform-by-platform settings in detail.
Step 2: Use Technical Protection Tools
Several tools now exist specifically to poison the well for AI training:
- Nightshade and Glaze — free tools from the University of Chicago that add invisible perturbations to your images. To human eyes, the photo looks unchanged; to AI training systems, the image is corrupted — it teaches the model wrong associations. Artists use these widely, and they work for personal photos too. The catch: you must apply them before uploading.
- Metadata stripping — remove EXIF data (location, device, timestamps) before posting. Most platforms strip some metadata anyway, but don’t rely on it.
- Watermarks — visible watermarks won’t stop determined scrapers, but they degrade the training value of images and assert your ownership.
Step 3: Control Where Your Photos Live
- Prefer private sharing. Family photo albums shared via private links (Google Photos shared albums, iCloud shared albums) are far less exposed than public social posts.
- Be cautious with photo apps. Read the terms of “free” photo editing, backup, or AI-enhancement apps — some claim broad rights over uploaded images, including for AI training. If the app is free, your photos may be the product.
- Check cloud service terms. Major providers (Google, Apple, Microsoft) generally don’t train AI on your private cloud photos — but verify the current terms rather than assuming.
- Think before posting children’s photos. Kids can’t consent to having their likeness in training datasets that will exist for decades. Many privacy advocates now recommend minimising children’s public photo footprint entirely.
Step 4: Exercise Your Legal Rights (UK)
Under UK GDPR, you have rights that apply here:
- Right to object to processing of your personal data for AI training — platforms operating in the UK must provide a way to object.
- Right to erasure — request deletion of your data from datasets and services where feasible.
- Copyright — your photos are your copyrighted works. Commercial AI training on them without permission is legally contested territory, and the landscape is shifting in creators’ favour.
The Information Commissioner’s Office has taken an active interest in AI training data practices — its guidance and enforcement actions are worth following if this matters to you.
What About Photos Already Scraped?
Be realistic: major datasets have already been built, and existing models can’t “unlearn” your photos easily (machine unlearning is an active research area, not a deployed solution). Focus your energy on:
- Preventing future scraping through the steps above.
- Supporting the legal and regulatory push for opt-in consent models — the policy battle matters more than any individual technical fix.
- Knowing your rights if your likeness is misused — non-consensual AI imagery of real people is increasingly covered by UK law (see our coverage of AI image risks).
Children’s Photos Deserve Extra Thought
Of all the photos online, children’s deserve the most careful thought. A child whose photos train AI models today will live with that data footprint for 80+ years — across technologies and norms we can’t predict. They can’t consent, and “sharenting” decisions made casually now are effectively permanent. This isn’t about fear; it’s about proportionality. Consider: does this photo need to be public, or would a private shared album serve the same purpose for grandparents?
Practical steps for family photos: default to private sharing (shared albums with links, not public posts); avoid posting identifying combinations (name + school + location); use platform audience controls per-post, not just account-wide; and periodically audit what’s public. Some families adopt a simple rule — no faces of children on public accounts — which sounds strict until you consider the alternative is a permanent, searchable, AI-trainable record of your child’s entire childhood.
For professional photographers, the considerations extend to clients: update your contracts to address AI training explicitly — who owns the images, whether they may be used for AI training, and what happens if a platform does so. Clients increasingly ask about this; having a clear answer is becoming a competitive advantage. The direction of travel is clear: consent norms are tightening, regulation is coming, and the photographers and families who acted early will be glad they did.
The Bottom Line
You can’t fully stop AI from using photos already scraped, but you can sharply reduce future exposure: lock social accounts to private and use platform opt-outs, apply protective tools like Nightshade before uploading, share privately rather than publicly (especially children’s photos), scrutinise photo apps’ terms, and exercise your UK GDPR rights. No single step is a complete shield — but together, they move you from “open buffet” to “hard target.” In the AI training data economy, friction is protection.