Image Metadata & Provenance Check

The verify-image-metadata pipeline is the fastest way to assess an image's authenticity. It interrogates the embedded EXIF metadata and the low-level file format for signs that the image has been altered, stripped of provenance, or fabricated. Common use cases include press photo triage, evidence intake screening, and bulk image review in investigative workflows where you need a reliable first signal before committing to a more resource-intensive analysis.

Unlike pixel-level deepfake detection, this pipeline focuses on what the file itself says about its own history: where it was taken, what device was used, what software last touched it, and whether the format is internally consistent. These signals are fast to extract, deterministic, and highly informative for the majority of everyday manipulation cases.

Pipeline Details

Property	Value
Pipeline ID	`verify-image-metadata`
Tier	1 — AI-Only
Estimated Duration	~1–3 min
Credit Cost	900 credits per run
Crowd Stages	No
Files Required	Yes — enriched image (JPEG, PNG, WebP, TIFF, or GIF)

Required Inputs

Files

One enriched image file. The file must have completed the enrichment step before the run is created (i.e. enrichmentStatus must be "completed"). See Files & Enrichment for details.

Context Keys

Context keys are optional for this pipeline but significantly improve the quality of the verdict when provided. The AI uses them to cross-reference embedded metadata against claimed provenance.

Key	Required	Description
`claimed_source`	Optional	Claimed source or publication where the image originated (e.g. `"AP Photo Agency"`, `"Reuters"`).
`claimed_date`	Optional	Claimed capture or publication date in ISO 8601 format (e.g. `"2024-11-10"`). Compared against EXIF creation timestamp.

Stages

EXIF Analysis (AI)

Extracts and evaluates the full EXIF metadata block embedded in the image file. The AI reviews GPS coordinates, camera model, device firmware, creation timestamp, and software fields. It checks for signs of EXIF stripping (where metadata has been removed, possibly to hide provenance), EXIF injection (where metadata has been synthetically added to a file that did not originally contain it), and internal inconsistencies — for example, a GPS location that contradicts a claimed date, or a camera model whose firmware version did not exist until years after the claimed capture date.

If claimed_source or claimed_date context keys are provided, the AI cross-references those claims against the embedded metadata and flags discrepancies.

Format Analysis (AI)

Inspects the image file at the format and compression level. The AI examines the file header, encoding parameters, compression artefacts, and pixel-level integrity to detect recompression or format conversion. Images that have been saved multiple times (for example, screenshotted from another screen, or re-exported after editing in Photoshop) typically show characteristic artefact patterns at compression boundaries. The AI assesses whether these patterns are consistent with the claimed origin and media type.

Verdict Synthesis (AI)

Aggregates the findings from the EXIF and format stages into a single final verdict. The AI weighs the relative strength of each signal — a missing EXIF block is weaker evidence of manipulation than an injected GPS coordinate — and produces a confidence-weighted verdict along with a human-readable summary explaining which signals contributed most to the outcome.

Starting a Run

curl -X POST https://api.crowdee.ai/v2/projects/{projectId}/verification-runs \
  -H "X-API-Key: crw_YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "pipeline_id": "verify-image-metadata",
    "file_ids": ["{fileId}"],
    "context": {
      "claimed_source": "AP Photo Agency",
      "claimed_date": "2024-11-10"
    }
  }'

Example Response

{
  "id": "run_01j9x...",
  "status": "completed",
  "verdict": "authentic",
  "confidence": 87,
  "pipeline_id": "verify-image-metadata",
  "scorecard": {
    "exif_consistency": "pass",
    "format_integrity": "pass",
    "claimed_date_match": "pass"
  },
  "explanation": "EXIF metadata is intact and internally consistent. Camera model and firmware are plausible for the claimed date. No signs of recompression or format conversion detected."
}

For deeper image analysis including crowd-sourced visual review and expert authentication — particularly useful when the image may be an AI-generated deepfake or a sophisticated composite — see verify-image-deep.