Video Technical Integrity Check

AI analysis of video container, frame consistency, and audio-visual synchronisation.

The verify-video-technical pipeline examines a video file at the technical level to identify signs of manipulation, editing, or format inconsistency. It inspects the container metadata and codec parameters, analyses extracted frames for visual discontinuities, and checks whether the audio and video tracks are temporally consistent with each other. This pipeline is designed for fast intake screening — press footage triage, evidence intake, or social media monitoring — where you need a reliable automated signal within minutes rather than hours.

The pipeline does not attempt to identify faces, detect specific individuals, or perform deep semantic analysis of the video content. Its strength is in the technical layer: detecting the artefacts that editing and re-encoding inevitably leave behind.

Pipeline Details

Property	Value
Pipeline ID	`verify-video-technical`
Tier	1 — AI-Only
Estimated Duration	~1–3 min
Credit Cost	1,200 credits per run
Crowd Stages	No
Files Required	Yes — enriched video with extracted frames (MP4, MOV, WebM, or AVI)

Required Inputs

Files

One enriched video file. The enrichment step extracts key frames as separate image file records (available in metadata.extractedFrameFileIds) and analyses the audio track. Both the extracted frames and audio track metadata are used by this pipeline. The file must have enrichmentStatus: "completed" before the run is created. See Files & Enrichment for details.

Context Keys

All context keys are optional. Providing them allows the AI to cross-reference claimed context against the technical signals it finds in the file.

Key	Required	Description
`claimed_source`	Optional	Claimed source or outlet where the footage originated (e.g. `"BBC News"`, `"Reuters"`).
`claimed_date`	Optional	Claimed recording or publication date in ISO 8601 format (e.g. `"2024-09-14"`).
`claimed_location`	Optional	Claimed filming location (e.g. `"Kyiv, Ukraine"`). Cross-referenced against any GPS data embedded in container metadata.

Stages

Container Analysis (AI)

Inspects the video container metadata, codec information, and encoding parameters. The AI reviews the creation timestamp, encoder software, muxer version, and any GPS or location tags embedded in the container. It checks for inconsistencies between the declared encoding parameters and the actual bitstream properties — for example, a file claiming to be a direct camera export but showing encoding settings only produced by re-encoding software. Mismatches between container-level and stream-level timestamps are also flagged.

Frame Analysis (AI)

Reviews the extracted key frames for temporal discontinuities, compression artefact patterns, and frame-level manipulation signs. The AI checks whether the compression artefact signature is consistent across frames — sudden changes in artefact intensity can indicate that frames from a different source have been inserted. It also assesses whether colour grading, lens distortion, and noise characteristics remain consistent throughout the clip, or whether there are visual breaks that suggest a cut-and-splice edit.

Audio-Visual Sync (AI)

Checks the temporal alignment between the audio track and the video track. The AI evaluates lip-sync consistency in any segments containing visible speech, and assesses whether the audio-visual temporal relationship is consistent throughout the clip. Desynchronisation — particularly when it appears or disappears mid-clip rather than being uniformly present — is a strong indicator that the audio or video has been replaced or shifted.

Verdict Synthesis (AI)

Aggregates all container, frame, and sync signals into a single final verdict. The AI weights each signal by confidence and produces a verdict with a structured scorecard and a human-readable explanation of the most significant findings.

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-video-technical",
    "file_ids": ["{fileId}"],
    "context": {
      "claimed_source": "BBC News",
      "claimed_location": "Kyiv, Ukraine"
    }
  }'

Example Response

{
  "id": "run_01j9x...",
  "status": "completed",
  "verdict": "manipulated",
  "confidence": 91,
  "pipeline_id": "verify-video-technical",
  "scorecard": {
    "container_consistency": "pass",
    "frame_artefact_consistency": "fail",
    "audio_visual_sync": "warn"
  },
  "explanation": "Frame analysis detected a significant change in compression artefact signature at 00:01:43, suggesting frames from a different source were inserted at that timestamp. Audio-visual sync shows minor drift in the same region, consistent with a cut-and-replace edit."
}

For comprehensive deepfake detection with crowd visual review and expert sign-off — including identity verification of people visible in the footage — see verify-video-full.