Every physical camera has a unique sensor fingerprint known as PRNU. DotShield compares the image’s noise residual to known camera signatures and to synthetic noise patterns produced by modern diffusion models.

• Matches frames to enrolled cameras using PRNU fingerprints.
• Detects uniform synthetic noise typical of AI-generated images.
• Evaluates shadow vs highlight noise — real sensors and AI behave very differently.

DotShield inspects pixel-level structure for JPEG artefacts, chromatic aberration, resampling and AI upscaling traces. Clean, “too perfect” pixels across the whole frame are often a bigger red flag than obvious glitches.

• Analyses JPEG/DCT block patterns and recompression signatures.
• Checks for realistic lens behaviour, colour fringing and edge softness.
• Spots AI super-resolution artefacts and beautification filters.

DotShield checks whether the world inside the image obeys physical rules: mirror behaviour, depth, shadows, perspective lines, text distortion and object repetition. A mirror selfie with forward-facing logos, for example, is an instant AI indicator.

• Validates shadow direction and softness against light sources.
• Tests mirror text and logo orientation for physical realism.
• Detects repeated textures or “too regular” object layouts.

Using the same depth and motion logic that powers DotShield’s liveness engine, the forensic pipeline checks faces, hands and posture for signs of replay, static masks or anatomical anomalies that arise in synthetic portraits.

• Analyses micro-motion, depth cues and parallax around the subject.
• Evaluates hand and finger geometry, jewellery behaviour and fabric stretch.
• Optionally correlates with prior trusted DotShield sessions for identity continuity.

DotShield’s Prompt Reconstruction Engine is a new class of forensic tool. It reverse-engineers “Prompt DNA” — the latent blueprint used by a generator to create a synthetic scene — without needing the original text.

• Extracts semantic tokens (objects, clothing, room layout, style).
• Detects structured generator schemas (e.g. “mirror_rules”, “preserve_face”).
• Links multiple fake images to the same operator or template through Prompt DNA similarity.

Instead of just saying “this looks fake”, DotShield can say: “This image matches a lifestyle selfie template with preserved face, iced drink, bedroom aesthetic, and overridden mirror physics.”