Optics and perspective
Progressive defocus, aberrations, vanishing lines, plane of focus. A real focal length behaves in a precise way: when it does not add up, that is the first crack.
About us
The most advanced image decoding platform in the world
The analyses available today read the source code of a photo and say, more or less, "yes" or "no".
But a generated image that is photographed, scanned or captured as a picture in picture loses that code — and if it is well made it stays indistinguishable to those checks. What was missing was something that reads an image the way a trained eye would: in the matter, in the light, in the shadow.
So why not bring together the finest photographers who ever lived and have them reason as one? Uthentic — built on the proprietary GuastoniUnmask code — runs multiple engines that question each other, each specialised in one layer of the photograph: matter, skin, light, texture, optics, anatomy, file data. The result is a layered, deterministic report, not a number to be taken on trust.
The layers of reading an image
Matter and grain
Skin texture, fabrics, surfaces. Real skin has an irregular microtexture; synthetic skin tends to melt into a uniform, almost waxy visual paste. GuastoniUnmask reads exactly this layer: skin microtexture and the coherence of pigmentation backlighting.
Light and capture
An identifiable or diffused source, layered or flat shadows, coherent catchlights in the eyes. GuastoniUnmask checks that the backlighting of skin pigmentation matches the reflected light of the environment; bounces that are too uniform are one of the signatures generative models betray most.
Anatomy and geometry
Hands, fingers, joints, the perspective of a face: this is where AI still stumbles.
Pixel structure and frequencies
The microscopic fingerprints left by GAN and diffusion models, invisible to the naked eye.
Local file integrity
It separates a real but retouched photo — cut and paste, cloning, object removal — from an image born synthetic. Two different things, which Uthentic never confuses.
Beyond the verdict: gear, re-capture, attribution
Uthentic estimates the plausible gear — camera, lens, shooting settings — with a declared attributability index. It tests the re-capture hypothesis: prints or screens re-photographed to wash away the traces of the first generation. And when an image is generated, it indicates which model family probably produced it — Midjourney, GPT-Image, Stable Diffusion — motivating every hypothesis.
Every signal, every suspicious area, every engine involved flows into a calibrated truth rate, with the margin of error and the level of agreement between models stated openly. When the evidence is contradictory, Uthentic says so instead of forcing a reassuring number. The fusion of engines is validated on a repeatable bench of 10,179 runs, with a category dedicated to real photographs worked in post production.
Where it really matters
- Journalism
- Insurance
- Media
- Publishing
- Legal and forensic evidence
Wherever you need to certify what is real before it becomes a decision, a news story or evidence in court. Uthentic delivers all of this as a forensic PDF analysis, with a QR code and sampling details.