{
  "name": "scene-window-W02-car",
  "model": "gemini-3-flash",
  "transport": "inline_data (base64 bytes of the actual MP4 incl. audio)",
  "endpoint": "https://api.kie.ai/gemini/v1/models/gemini-3-flash-v1betamodels:streamGenerateContent",
  "prompt": "This clip is an ORIGINAL excerpt with sound from a longer film. It covers source time 12.00–30.00 s (clip 0.00 = source 12.00). Clip time runs at normal speed. Watch and listen to all of it; answer only from what is visible/audible in THIS clip.\n\nGoal: a conversation-level ledger of how meaning is built between people — who speaks, who listens, what each listener does while listening, when a reaction starts relative to the words, what a gesture or object does, and what each cut carries forward. This is evidence gathering, not evaluation.\n\nReturn ONLY valid JSON (no markdown fences):\n{\n \"access\": {\"heard_audio\": bool, \"one_sound_detail\": str, \"one_picture_detail\": str},\n \"utterances\": [{\"clip_start_s\", \"clip_end_s\", \"speaker\": role, \"text\": exact words, \"addressed_to\": role, \"delivery\": pace/pitch/volume/breath/pauses in a few words, \"on_screen_while_speaking\": who is framed, \"listener_reaction\": what a visible listener does (gaze, brow, mouth, head, hands, body) and roughly when relative to the words (before/during/after which word), \"pause_after_s\": estimated silence after the line}],\n \"shots\": [{\"clip_start_s\", \"clip_end_s\", \"scale\", \"framing\": who/what and eyeline/angle, \"action_at_start\": first visible action, \"action_at_end\": last visible action, \"hands_and_props\": what hands and objects do, \"gaze\": where each visible person looks, \"cut_out_reason\": what the next shot needs to show that this one cannot (your best reading), \"continuity_into_next\": does a movement/sound continue across the cut?}],\n \"events_inside_shots\": [meaningful changes inside a continuous shot: {\"clip_time_s\", \"who\", \"what_changes\": observable, \"trigger\": the word/sound/action that immediately precedes it, \"confidence\"}],\n \"information_flow\": [{\"clip_time_s\", \"who_learns_what\", \"from_whom\", \"visible_evidence_of_receipt\"}],\n \"sound\": {\"ambience\": str, \"music\": \"none|present\", \"silences\": [clip times], \"sound_across_cuts\": notes on whether speech/ambience bridges a cut},\n \"uncertainties\": [str]\n}\nTimestamps are approximate; do not claim frame precision. Do not invent reactions that you cannot see.\n",
  "prompt_sha256": "c6dd8fdb82c98ba41517f732bfb1126f9457efba6f1547091649272b59c420bb",
  "media": {
    "path": "/workspaces/UGS/AdFactoryNarrativeFormat/trees/bf8f11bde4/analysis/original-atomic/fragments/scene-windows/W02-car.mp4",
    "sha256": "bd407dc68e3fb37f7664dab64444e01c4c6f3e71081b1e87480454594be03078",
    "bytes": 1554855,
    "mime": "video/mp4"
  },
  "extra_media": [],
  "generation_config": {
    "thinkingConfig": {
      "includeThoughts": false,
      "thinkingLevel": "high"
    },
    "maxOutputTokens": 30000,
    "temperature": 0.2,
    "mediaResolution": "MEDIA_RESOLUTION_HIGH"
  },
  "meta": {
    "window": {
      "id": "W02-car",
      "start_s": 12.0,
      "end_s": 30.0
    },
    "clip": {
      "path": "/workspaces/UGS/AdFactoryNarrativeFormat/trees/bf8f11bde4/analysis/original-atomic/fragments/scene-windows/W02-car.mp4",
      "sha256": "bd407dc68e3fb37f7664dab64444e01c4c6f3e71081b1e87480454594be03078",
      "bytes": 1554855,
      "source_start_s": 12.0,
      "source_end_s": 30.0,
      "duration_s": 18.0,
      "recipe": [
        "ffmpeg",
        "-hide_banner",
        "-loglevel",
        "error",
        "-y",
        "-ss",
        "12.000",
        "-to",
        "30.000",
        "-i",
        "/workspaces/UGS/AdFactoryNarrativeFormat/data/references_liked/resilia-aged-garlic-family-cholesterol-story.mp4",
        "-vf",
        "scale=-2:854",
        "-c:v",
        "libx264",
        "-preset",
        "medium",
        "-crf",
        "24",
        "-pix_fmt",
        "yuv420p",
        "-c:a",
        "aac",
        "-b:a",
        "96k",
        "-movflags",
        "+faststart",
        "/workspaces/UGS/AdFactoryNarrativeFormat/trees/bf8f11bde4/analysis/original-atomic/fragments/scene-windows/W02-car.mp4"
      ],
      "time_scale": 1.0
    },
    "source_start_s": 12.0,
    "time_scale": 1.0
  }
}