How to do it
Limiting the use of your photos isn't done with a single click, but with a few moves that together greatly lower the risk.
- Look for the opt-out option from training. Several platforms have added, in their privacy settings, an item to not use your content to train their AI. Look for it as "AI training," "use of data for improvement" or similar. Be aware: in some geographic areas the option exists, in others it doesn't, and whether it's available depends on local rules.
- Reduce your public surface. Every public photo can be collected. Set your profiles to private, limit the audience of your posts, and publish fewer high-resolution frontal close-ups, which are the most useful material for those who want to recreate a face.
- Use "masking" tools for the images you want to protect. There are free programs, developed by university research groups, that apply alterations imperceptible to the human eye but that confuse the models trying to imitate a face or a style. They're designed mainly for artists and creators, but they work for any image you care about. Look for them as "Glaze" and "Nightshade."
- Check the terms of the services you use. When you upload photos to an app, you accept its terms: some reserve broad usage rights. Before entrusting important images to a new service, take a look at what it says about uploaded content.
Check: you've done the part that depends on you when you've enabled the available opt-out options, restricted public visibility, and protected with masking tools the few images you truly hold dear.
A concrete example
Giorgia is an illustrator and fears her style and her face will be copied. She doesn't just hope. In the settings of the platforms she uses she enables, where available, the exclusion of her content from training. She sets her personal profile to private and keeps only her professional one public, with low-resolution images. For the works she cares about most, before publishing them she runs them through a free masking tool: to the eye they're identical, but for a model trying to imitate them they come out "disturbed." She doesn't have absolute certainty, but she has raised the wall as high as possible with free tools.
When it does NOT work (and how to fix it)
If the platform doesn't offer the opt-out option
Not all companies let you refuse the use of your data, and in some countries the option isn't provided. Where the switch is missing, you still have control over what you upload: if a service reserves the right to use your content as it likes and you can't object, the defense is to publish less sensitive material there, or not publish it at all.
If your photos have already been public for a while
What's already been online may have been collected, and opt-outs usually apply to the future, not the past. The realistic remedy is to act on the present: make private or remove what you can, enable the opt-outs from now on, and mask important images before republishing them. Reducing future availability is valuable anyway.
If someone uses your image without permission
If you discover an unauthorized use of your face, you have some footholds. Report the content to the platform that hosts it (the removal of images portraying a person without consent is provided for by many services), keep the evidence, and in serious cases consider a legal route: various legal systems protect the right to one's own image and, increasingly often, to artificial reproductions of face and voice.
A tip from someone who really uses it
Treat your face and your voice like a password you can't change. A compromised password you change; your face, no. That's why it pays to be selective about what you make public and high-definition: not out of paranoia, but because what goes out can't be recalled. Publish with the awareness that every sharp image of your face is a building block that someone, one day, could use.
Frequently asked questions
Do masking tools ruin my photos?
No, they're designed specifically not to: the alterations they apply are engineered to stay imperceptible to the human eye, while confusing the models. The photo stays normal for whoever looks at it. They aren't infallible, though, and don't guarantee eternal protection, because training systems evolve: they're a serious obstacle, not a definitive shield.
If I enable the opt-out from training, do companies respect it?
Serious companies apply the options they offer, but not all offer the same thing, and the real degree of respect depends on the rules and their good faith. Enabling the opt-out is right and should be done, but don't fool yourself that it's an absolute guarantee: that's why it counts to combine it with reducing what you publish, the only lever entirely in your hands.