DeadDetect API

Verification API that returns UID, confidence, and evidence for backend workflows.

What is DeadDetect

DeadDetect is the verification layer: upload media (or select from your library) and check whether a DeadMark payload is present.

It returns a practical result model: present / uid / confidence / evidence.

How it works

Two modes keep this practical:

Normal Scan: fast checks for day-to-day moderation and monitoring.

Deep Scan: used when media is heavily transformed and UID recovery is weak. It compares recovered signal traits and media parameters against internal encode/decode logs, then returns an attribution confidence score.

C2PA is used when available, but verification does not depend on C2PA alone.

Active development: Deep Scan no-watermark AI-origin estimation is in progress, including media model-family attribution outputs for image, video, and audio.

Media evidence preview

DeadDetect runs in backend/moderation pipelines, but escalated cases still need clear evidence. The image, video, and audio patterns below show how that review output can look in internal tools.

C2PA manifest traceDeadMark UID watermarkInternal decode logs

Original

Slide to reveal the watermarked image

Detection overlay (demo UI)

Watermark detected

UIDDMK-IMG-9A72-1F3C

Confidence98.6%

Integrity health96 / 100

C2PA manifestRecovered

Decode logApp log + timestamp chain

Trace + provenance

C2PA manifestIntegrated trace

DeadMark UID watermarkPrimary signal

Decode logs + timestampsInternal audit trail

If C2PA is stripped, UID + internal logs still support source verification workflows.

Example generation + detection metadata

Model: FLUX.1 Pro (demo metadata label)

Prompt: "Mountain valley with snowy peak, pine forest foreground, cinematic daylight realism."

Generated by: @demo_creator_17 * IP 198.51.100.24

Detected on: Instagram repost by @sample_repost_account * 2026-02-24 18:41 UTC

Dots only show evidence locations in the UI. They are not visible watermark pixels. Placement and strength are ML-adaptive per asset.

Workflow note: C2PA + internal verification

We use C2PA when available and pair it with UID watermark + decode logs, so verification is not dependent on metadata alone.

Roadmap signal

Text/document/code workflows were started early and are now shown in preview UI. Deep Scan no-watermark AI-origin is in active development.

Text + code preview

Text, docs, and code watermarking

DeadMark is expanding beyond image/video/audio with one API and format-aware profiles for text workflows. The target is practical survivability through copy/paste, spacing changes, and light reformatting.

Text detection preview

ChatGPT sample (demo)

"Daily operations update: all three service regions are stable after maintenance. Throughput is back to normal and alert volume remains low."

Detection overlay: UID DMK-TXT-7D1A-29F4 * confidence 96.8% * integrity 92 / 100 * recovery: spacing-normalized path.

Messaging

Chat apps, short posts, comments

Plain text

TXT, Markdown, structured text

Documents

DOCX, PDF text layers, exports

Code

Repositories, snippets, generated files

Code detection preview

Claude Code sample (demo)

export function scoreRisk(input: string): number {
  const normalized = normalizeText(input);
  const entropy = estimateEntropy(normalized);
  return Math.min(100, Math.round(entropy * 37));
}

Detection overlay: UID DMK-CODE-11C8-5AE2 * confidence 97.2% * integrity 94 / 100 * recovery: structure-preserved path.

Output model: present / uid / confidence / recovery status.

AI-origin estimation (no watermark, media) - Deep Scan in development

Model-signal attribution

Media-generation signal hints

Diffusion residual texture patternHigh

Temporal synthesis cadence patternMedium

Generative audio texture patternMedium

Model confidence

DALL-E family82%

Kling family74%

Seedance family69%

Deep Scan no-watermark attribution is media-focused (image, video, audio). This block is model-family confidence output, not watermark decode output.

Scan modes

Normal Scan

Fast and lightweight
Best for routine checks
Returns UID + confidence + evidence summary when decode is healthy

Deep Scan

Used when watermark signal is severely damaged or transformed
If UID is unrecoverable, compares media traits to internal logs and known parameters
Outputs attribution confidence score and evidence package
No-watermark AI-origin estimation is currently in active development

FAQs

You can, but you will not get UID-linked evidence unless content was encoded. For generic AI-origin classification, use a separate detector model.