AI image provenance: what it is
Quick facts
- Definition
- Information about the origin of an image: who made it, how it was generated and whether it was altered.
- Why it matters
- Prevents an AI image from being mistaken for a real photograph or a patient result.
- Tools
- Visible notice on the image, caption, internal record and, where available, authenticity metadata.
- Rule of thumb
- An AI-generated image must always be identified as a simulation.
The provenance of an AI image is information about the origin of an image created or modified by artificial intelligence: who produced it, with which tool, from which photo and whether it was edited afterward. In a world where synthetic images are increasingly convincing, provenance is what makes it possible to tell a real record from a simulation.
Why does it matter in HOF?
The clinic works with images. The patient's photo is a clinical record, and the projection of the result is a conversation tool. Mixing the two can mislead the patient, the public and the regulators.
| Type of image | What it is | How to handle it |
|---|---|---|
| Clinical photograph | Real record of the patient | Patient record, with consent |
| AI simulation | Projection of what may be pursued | Always identify as an illustrative simulation |
| Fully synthetic image | A face that does not exist | Identify as generated by generative AI |
| Manipulated image | Altered real photo | Do not use as a result |
The central risk is misleading advertising: presenting a simulation as a real result leads the consumer into error. Checking the rules of your professional council and of CONAR on advertising is part of the routine.
What practical precautions should you take?
- Label: every simulation must carry a visible notice that it is illustrative.
- Separate: projection files and real photographs in different folders, with clear names.
- Do not mix in marketing: never post a projection as a real "before and after".
- Consent: the patient needs to know their photo will be processed by AI, that the generated image is an estimate and how the data is handled. See informed consent and LGPD (Brazil's data protection law).
- Sensitive data: a face photo is facial biometric data when used for identification, and deserves stronger protection. See anonymization.
- Limits of AI: models can invent details. See hallucination in image AI and inpainting.
How does Harmy handle this?
The app produces a projection that changes only what was marked, and so it keeps the person recognizable and the image honest. Even so, the responsibility for presenting it as a simulation remains with whoever shows it.
Also called: origin of synthetic image, image provenance.
Frequently asked questions
Why identify an image as AI-generated?
So as not to lead the patient or the public to believe it is a real result. Identification protects transparency, ethics and the professional in the face of any questioning.
Can a simulation be used in an advertisement?
Not as if it were a result. A simulation is illustrative. Check the rules of your professional council and of CONAR (Brazil's advertising self-regulation body) on advertising, and never present a projection as a real before and after.