YouTube, Instagram and TikTok Video Evidence Compiler

Compile public YouTube, Instagram, and TikTok videos into timestamped transcript evidence or an ordered tutorial draft.

Accepted input

videoUrls accepts public YouTube, Instagram, and TikTok post URLs, YouTube video IDs, and direct public media URLs. Select videoEvidence to locate exact spoken phrases or tutorialToSop to preserve spoken tutorial segments as draft steps.

{
  "scraperType": "videoEvidence",
  "videoUrls": [
    "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
    "https://www.instagram.com/reel/example/"
  ],
  "evidenceTerms": ["turn off power", "remove the cover"],
  "language": "en",
  "detectScenes": true,
  "sceneThreshold": 27,
  "includeTranscriptText": true,
  "maxItems": 20
}

Matching is literal after case and punctuation normalization.

Analysis modes

videoEvidence returns the spoken opening found within the first three seconds, exact selected phrase matches, word and segment timestamps, and the scene indexes overlapping each piece of evidence.

tutorialToSop returns one ordered draft step per usable spoken transcript segment, up to maxSopSteps. Each step keeps the spoken instruction, timestamps, transcript segment ID, and overlapping scene indexes. The draft preserves spoken evidence; it is not a verified procedure.

Response fields

Each row represents one submitted video:

• video_source contains the platform, public post URL, and resolved public metadata.
• analysis_status is complete, partial, not_found, or source_error.
• media contains observed duration, size, format, dimensions, and audio/video availability.
• transcript contains language and coverage fields, plus text, segments, and words when enabled.
• scenes contains detected cut timings and motion labels.
• video_evidence contains the opening hook, exact matches, and the timestamped transcript timeline.
• sop_draft contains the ordered spoken steps in tutorial mode.
• analysis and provenance record which stages completed.

{
  "record_type": "video_evidence",
  "video_source": {
    "platform": "youtube",
    "url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
    "id": "dQw4w9WgXcQ",
    "title": "Example repair guide",
    "author": "Fix Lab"
  },
  "analysis_status": "complete",
  "video_evidence": {
    "hook": {
      "start": 0,
      "end": 2.4,
      "text": "First, turn off power."
    },
    "exact_matches": [
      {
        "phrase": "turn off power",
        "start": 0.7,
        "end": 2.4,
        "quote": "turn off power.",
        "segment_ids": [0],
        "scene_indexes": [0, 1]
      }
    ]
  },
  "analysis": {
    "resolve": { "status": "ok" },
    "transcribe": { "status": "ok" },
    "scenes": { "status": "ok" },
    "compile": { "status": "ok" }
  }
}

Behaviour on partial results

A video remains in the dataset when one selected stage fails. For example, successful scene detection with failed transcription produces partial, empty transcript evidence, and the scene timeline. An invalid source or a post with no playable media produces source_error; a resolved post with no record produces not_found.

Media analysis is limited to two minutes and 100 MB per video. Scene labels describe cuts and broad motion; they are not a substitute for visual interpretation.

Direct media URLs must be publicly reachable.