Inside Eterna Image Immunization: How Image Privacy Can Be Designed Before Misuse Happens

Most of what's written about deepfake defense focuses on catching misuse after it happens; this is about the earlier question, what makes an ordinary photograph usable as raw material for AI identity replication in the first place, and what it means to design against that before an image is ever published.

Why an ordinary photograph is a machine-readable identity source

To a person, a photo is a picture. To a modern image or generative model, it's also a set of extractable signals: facial geometry, distinguishing features, and patterns a model can learn to recognize or reproduce. Publishing an image publicly doesn't just share how someone looks with other people; it can also hand that same information to systems designed to learn from it. That gap, between what an image communicates to a person and what it exposes to a machine, is the starting point for Image Immunization.

Human-visible appearance versus machine identity utility

These are two different properties of the same image, and they don't have to move together. Human-visible appearance is what a person sees when they look at the photo. Machine identity utility is how usable that image is as a source for a system trying to extract or reproduce an identity from it. EIP is designed to work on the second property while leaving the first alone, so the image still looks like an ordinary photograph to anyone who sees it.

Pre-publication protection

EIP is applied before an image is published or distributed, at the point its owner authorizes it for use. That timing is deliberate: protection applied after an image is already circulating has to work against every copy, repost and derivative that followed it. Protection applied at the source only has to be present once, at the point of publication.

Why proactive protection complements detection and enforcement

EIP does not replace Eterna's detection, monitoring and response services, and it isn't designed to. Detection and enforcement remain necessary for anything that happens after an image is already public: misuse of images that predate protection, content shared outside Eterna's reach, and cases that need evidence, platform reporting and human judgment. EIP is an earlier layer aimed at reducing how much of that work is needed in the first place, not a substitute for it.

Why no single protective layer should be treated as universal

No image protection technique, including EIP, should be described as working against every AI system, every transformation, or every future model. Systems evolve, and an image can be copied, cropped, recompressed or otherwise altered in ways that affect any protective technique differently. EIP is currently undergoing continued validation, and protection effectiveness may vary across systems, transformations and use cases. That's a design constraint worth being honest about, not a footnote to bury. See How Eterna Validates Image Immunization Responsibly for the full account.

Frequently asked questions

Is Image Immunization available today? Launch timing and access details will be announced separately.

Does it guarantee an image can't be misused? No. No protective layer can guarantee that, and EIP does not claim to.

Will it change how my photo looks? No. EIP is designed to preserve the image's natural, human-visible appearance.

Does this replace Eterna's detection and response services? No. It's an additional layer, applied before publication, alongside them.