SIG-01
Metadata & Provenance
Reads EXIF fields, camera and software tags, timestamps, and embedded content credentials (C2PA) when present.
Provenance documents a file’s history — it doesn’t prove the depicted event is real.
Windows beta — now enrolling
Don’t just ask if it’s fake.
Examine the evidence.
SUS Detect is a forensic analysis tool for images and video. It inspects editing traces, manipulation signals, AI-generation evidence, provenance data, and internal inconsistencies — then shows you exactly what it found, where, and why.
Free during beta · Windows 10/11 · Evidence, not verdicts
Why it’s different
A detector gives you a verdict.
SUS Detect gives you evidence.
Single-score “AI detectors” compress a complicated question into one number you can’t check. SUS Detect takes the opposite approach: run multiple independent forensic checks, keep every signal separate, and show the underlying evidence — so the conclusion is yours to make, not the software’s to hide.
Multiple independent signals
No single point of failure. Each check runs on its own and reports on its own — a miss in one doesn’t silence the others.
Every finding is inspectable
Findings point back to the region, metadata field, or artifact that produced them. You can verify anything the report claims.
Honest about uncertainty
Experimental detectors are labeled experimental. Ambiguous results are shown as ambiguous. Weak evidence stays weak.
01 / How it works
From suspect file to evidence report
in three steps.
01
Add media
Start with an image or a video file — from a messaging app, a camera card, a download, or a submission you need to verify.
02
Run the analysis
SUS Detect applies its forensic checks: metadata and provenance inspection, compression and noise analysis, editing-trace detection, and experimental AI-generation signals.
03
Review the evidence
Findings are grouped by signal with the underlying evidence attached. Examine each one, then export the report to share with your team.
02 / What it examines
Six ways a file can betray itself.
Every check answers three questions: what was found, where, and how strong is the evidence. No signal is treated as proof on its own.
SIG-02
Compression & Save History
JPEG quantization and error-level analysis reveal re-saves and regions compressed differently from the rest of the image.
SIG-03
Noise & Sensor Consistency
Compares sensor-noise patterns across regions of the image. Inconsistencies can indicate compositing or splicing.
SIG-04
Editing & Duplication Traces
Searches for cloned or resampled regions and the edge inconsistencies cut-and-paste edits leave behind.
SIG-05 EXPERIMENTAL
AI-Generation Signals
Statistical detectors flag patterns associated with generated media, reported separately from classic forensic checks.
Expect false positives and false negatives. Treat as a lead, never a conclusion.
SIG-06
Video Frame Sampling
Extracts representative frames from video and runs image-level analysis across the sample.
Sampled-frame analysis — not comprehensive temporal or audio deepfake detection.
03 / Read before you trust any result
What SUS Detect can’t tell you.
Forensic tools earn trust by stating their limits. These are ours — and they apply to every tool in this space, whether or not they admit it.
-
A clean result does not prove a file is authentic. Absence of detected signals is not evidence of absence.
-
A flagged signal is evidence for further investigation — not automatic proof of manipulation.
-
Experimental detectors produce false positives and false negatives. They generate leads, not verdicts.
-
Provenance and metadata describe a file’s history. They can’t prove the depicted event itself actually happened.
-
Video analysis inspects sampled frames. It is not comprehensive temporal or audio deepfake detection.
-
We don’t publish accuracy percentages. Detector performance depends too heavily on content type for any single number to be honest.
If you need a guaranteed answer, no tool can give you one. If you need evidence, start here.
04 / Who it’s for
Built for people who verify before they trust.
Journalists & fact-checkers
Triage user-submitted media before it shapes a story.
Investigators
Document forensic observations alongside the rest of your evidence.
Researchers
Study manipulation techniques with inspectable, exportable signals.
Legal & compliance teams
Assess digital exhibits with findings you can trace to a source.
Content-verification teams
Add a second layer of scrutiny to trust-and-safety review queues.
Technically serious consumers
Check the viral image yourself instead of taking a screenshot’s word for it.
05 / Beta access
Join the Windows beta.
The core software is built and working. The beta exists to harden it against real-world media and real-world workflows — including controlled external compatibility testing against your own file sets.
WHAT YOU GET
- Full access to the beta, free of charge
- A direct feedback channel to the developers
- Your test cases shape the detector roadmap
WHAT WE ASK
- Run it against real files, not just demos
- Report false positives and missed detections
- Treat results as evidence, not verdicts
REQUIRES · WINDOWS 10/11 · 64-BIT
06 / FAQ
Straight answers.
Is SUS Detect an “AI detector”?
No. AI-generation detection is one experimental signal among several. SUS Detect is a forensic analysis tool: it combines metadata and provenance inspection, compression and noise analysis, and editing-trace detection into a single workflow, and reports each signal separately with its evidence.
How accurate is it?
We don’t quote a single accuracy number, because no single number is honest. Detector performance varies with content type, compression history, and editing technique. Instead of a score you can’t verify, SUS Detect shows you what it found and the evidence behind it, so you can judge the result in context.
Can a clean result prove a photo is real?
No — and any tool that claims otherwise is overstating its capabilities. A clean result means none of the checks found a suspicious signal. Skilled edits can evade individual detectors, and absence of evidence is not evidence of absence.
Does it analyze video?
Yes, with an important limit: SUS Detect extracts representative frames and analyzes them with its image-level checks. It is not comprehensive temporal analysis and does not perform audio deepfake detection.
What platforms are supported?
The beta runs on Windows 10 and 11 (64-bit). Other platforms are under consideration and will be prioritized based on beta demand.
What does the beta cost?
Nothing. Beta access is free in exchange for what actually helps: real-world testing and honest feedback, including false positives and missed detections.