Skip to documentation

Selected TEST1 graph · source checked

One frozen teacher. Three residual heads. Two routes.

The current detector records native image size, extracts one Tu et al. Expert 4 representation, and applies a fixed routed correction. This guide keeps its evidence and rights limits next to the technical details.

01frozen Expert 4
03residual heads
0–1review signal

Start here

In one minute

SynthFlag uses a frozen upstream Tu et al. Expert 4 SigLIP encoder and teacher head. Three project-trained lightweight heads correct that teacher margin. Native images at most 64 px use a specialist; larger images use a fixed two-head stack.

Example output0.78

Higher means a stronger AI-positive model signal.

01

What goes in?

One image, eight browser-sampled frames, or a CLI folder.

02

What happens?

One 384 px SigLIP view is scored through a native-size route.

03

What comes out?

A continuous score, plus traceable batch or frame metadata.

04

What does it prove?

Nothing alone. It is a review signal, not origin proof.

01 · Preserve native context

Record size, then create one SigLIP view

Released code

Routing uses the original longest side, so dimensions are recorded before RGB conversion and resize.

  • Decode and validate the image.
  • Record max(native width, native height).
  • Bicubic-resize the short edge to 384 px.
  • Center-crop to 384 square and normalize with mean/std 0.5.

Input contract

  • RGB pixels only
  • no EXIF or provenance metadata
  • no random inference augmentation
  • video audio and motion are not analyzed

02 · Frozen teacher + project heads

The selected graph

Upstream lineageReleased code

Expert 4 produces a pooled 1,152-dimensional feature and two teacher logits. Each residual head adds one learned scalar to the detached teacher margin.

  • Teacher: Tu et al. Expert 4, SigLIP So400M Patch14-384.
  • Margin: logit[1] - logit[0].
  • Head: LayerNorm → Linear(256) → GELU → Dropout → Linear(1).
  • 297,729 parameters per head; 429,414,469 loaded total.
corrected margin =teacher + α × residual(feature)

03 · Deterministic route

The exact score conversion

≤64 px route

CIFAKE specialist

score = sigmoid(teacher + 1.25 × residual_low)

>64 px route

Fixed two-head stack

Blend corrected epoch-05 and epoch-08 margins 0.65 / 0.35, then subtract -1.557959395647049 before sigmoid.

TEST1 reports threshold metrics at score 0.5. That convention is not a universal production threshold or calibration guarantee.

04 · Public development diagnostic

TEST1 results

CIFAKE AUC0.9816

clean · 0.9095 composite

SID-Set AUC0.8691

clean · 0.8439 composite

WildFake AUC0.9467

clean · 0.8785 composite

TEST1 contains 15,000 unique public sources and 30,000 aligned clean/composite evaluations. It is not TikTok's hidden test. The public suites were previously inspected and native resolution sends all TEST1 CIFAKE images to the specialist.

Research input · Day 3

Professor Ng Teck Khim: a future camera-forensics direction

The SynthFlag team interviewed Professor Ng Teck Khim about local image statistics created by Bayer color-filter sampling and demosaicing. The discussion motivated future tests of small-block variance, cross-channel correlation, mosaic phase, and acquisition consistency.

Interview-derived direction
Professor Ng Teck Khim and the SynthFlag team during a Day 3 video research interview
Team-supplied interview image. It documents the conversation, not detector performance or endorsement.

What the interview changed

  • Test whether physical-camera statistics complement deep features on the same source-disjoint images.
  • Stress blur, resize, JPEG, screenshots, crop, color processing, and deliberate Bayer-statistic injection.
  • Validate low-false-positive operating points rather than relying on ROC-AUC alone.
Download the interview-only transcript
Exploratory, not a final result

The early prototype was not strong or stable enough for a result. It was not used in the selected Expert 4 plus three-head runtime, TEST1, or final model selection. No performance metric is claimed for it.

05 · Keep the boundary attached

Limitations, rights, and eligibility

Owner accepted

Collaborator rights attestation

The collaborator attests that the residual heads and training inputs are rights-cleared for project use. The project owner accepts that assurance; it was not independently audited here.

Upstream dependency

Expert 4 remains upstream

Its redistribution permission is unproven, and an existing AIGC detector may require explicit organizer clearance.

Error profile

Both errors matter

SID misses many local tamper positives; composite WildFake produces 861 false positives at the reported boundary.

Interpretation

Use review and provenance

Do not infer creator, generator, edit location, or authentic provenance from one score.

06 · Product contract

Service, video frames, and batch records

API

One cached model

One active and one queued request; prediction is serialized.

VIDEO

Eight local samples

Only midpoint PNG frames cross the upload boundary.

CLI

Resumable output

CSV, atomic Track 5 JSON, and run metadata.

CHECKS

Four file identities

Expert 4 and all three heads pass size/hash verification.

Trace the evidence

Sources and downloads

The retired 18-diagram four-expert atlas remains in repository history as baseline documentation. It is not the selected runtime contract; current behavior is defined by the Expert 4 plus three-head implementation and manifest.