Was This Image
Tampered With?
The Styx Forensic Suite answers that question using physics, not pattern matching. It extracts a 284-dimensional forensic fingerprint from any image—identifying the specific manipulation applied, quantifying degradation severity, and tracing multi-stage re-compression chains. All without training data, and all without retraining when new attack methods emerge.
The Problem
Digital image integrity is under attack on every front. Deepfakes threaten identity, elections, journalism, and legal evidence. AI-generated content is flooding platforms faster than moderation can respond. And every image shared on social media passes through 3–5 stages of compression and resampling—destroying the forensic traces that traditional tools rely on.
The fundamental limitation of current approaches: deep learning forensic detectors are trained on specific manipulation types and fail on novel attacks. Statistical methods measure surface-level pixel properties but have no physical model of why distortions create specific signatures. Styx operates on a different principle entirely.
Current Forensic Tools
Trained on specific manipulation types — fail on novel attacks
Black-box neural networks — no explainability for courtrooms
Defeated by multi-stage re-compression chains
Require constant retraining as new AI generators emerge
Styx AI Approach
Physics-based — detects universal manipulation signatures
Fully interpretable — every feature has physical meaning
Resilient to multi-stage degradation chains
Zero retraining — works on attacks it has never seen
The Forensic Pipeline
Four Forensic Lenses
Every image is analyzed through four independent proprietary forensic perspectives. Each lens examines the deep spectral structure of the image—not raw pixel values—extracting signals that are invisible to conventional forensic tools and resilient to adversarial evasion.
Compression Forensics
Detects the telltale frequency-domain signatures left by JPEG compression, resolution scaling, and synthetic image generation. Can identify whether an image has been through one compression stage or five.
Manipulation Detection
Identifies directional texture anomalies introduced by editing tools—splicing, inpainting, clone-stamping, and AI-assisted retouching all leave distinct multi-scale disruption patterns that this lens captures.
Consistency Mapping
Fingerprints the physical coherence of image regions by measuring structural relationships between neighboring areas. Edited regions break these natural consistency patterns—providing court-admissible evidence of tampering.
Source Identification
Traces the origin and processing history of an image by analyzing its spectral energy distribution. Can distinguish camera-original, AI-generated, screen-captured, and multi-platform re-encoded images.
The Social Media Meatgrinder
Every time an image is shared on social media, the platform re-compresses, resamples, and color-shifts it. By the time an image has been shared across three platforms, it has been through 3–5 sequential degradation stages. Traditional forensic tools—which rely on detecting specific compression artifacts—lose the trail completely after the second stage.
Styx was specifically designed to solve this problem. Our proprietary "Meatgrinder" test protocol simulates exactly these multi-stage chains during development, and our four forensic lenses detect the cumulative spectral damage—even when individual artifacts have been overwritten by subsequent processing.
Blind Quality Scoring (Mjolnir Engine)
Every image receives a calibrated quality score from 0–100, with no reference image required. Unlike traditional metrics that compare against an original, Mjolnir measures the physics of degradation itself—quantifying how much spectral damage an image has accumulated regardless of its processing history.
By the Numbers
Every image produces a unique 284-dimensional forensic fingerprint—rich enough to identify the specific type and severity of manipulation, not just flag that something changed.
Blur, JPEG compression, noise, resampling, gamma distortion, color quantization, chromatic aberration, salt-and-pepper noise, motion blur, and color temperature shift—plus multi-stage combinations.
Core feature extraction is entirely physics-based. When a new deepfake generator or manipulation tool appears, Styx detects it immediately—no retraining, no dataset collection, no model update cycle.
Who It's For
Law Enforcement & Intelligence
Forensic labs at FBI, Interpol, and state crime labs need court-admissible evidence of image tampering. Styx provides fully interpretable, physics-grounded forensic reports—not black-box neural network outputs.
Legal & Insurance
E-discovery firms and insurance fraud investigators need to verify the authenticity of photographic evidence. Each Styx analysis produces a quantified integrity score backed by physical measurements.
Social Media Platforms
Content moderation at scale—Styx can screen images at upload time, flagging manipulated content, AI-generated media, and re-compressed misinformation before it spreads.
News & Media Verification
Wire services and newsrooms need to verify photo authenticity before publication. Styx provides provenance scoring that distinguishes camera-original content from AI-generated or manipulated imagery.
Why Physics Wins
Every deep learning forensic tool has the same fundamental weakness: it can only detect manipulations it was trained on. When a new AI generator appears—and they appear monthly—the entire model must be retrained on new data. Styx measures a universal physical signal: spectral incoherence. All image manipulations—regardless of method—disrupt the natural relationships between color channels. This disruption is governed by physics, not by the specific tool used to create it.
| Capability | DL Forensics | Statistical (BRISQUE) | Styx AI |
|---|---|---|---|
| Novel attack detection | ✗ Requires retraining | ○ Limited | ✓ Physics-based |
| Multi-stage chain detection | ✗ Fails after 2 stages | ✗ No capability | ✓ Built-in Meatgrinder |
| Distortion type identification | ○ Trained types only | ✗ No capability | ✓ 10 types + combos |
| Explainability | ✗ Black box | ○ Partial | ✓ Every feature interpretable |
| Training data required | Millions of labeled images | Pristine image corpus | None |
Get in Touch
The Styx Forensic BIQA Suite is built on original mathematics and proprietary algorithms developed from the ground up. The core engines are mature and validated — we're now seeking investment partners to fund continued R&D and bring this technology to market.
If you're interested in the future of physics-based image forensics — whether as an investor, research collaborator, or integration partner — we'd love to hear from you.