{
 "S68sh3": {
  "input_fingerprint": "a418e9509534618d",
  "geometry": {
   "entity_placements": [
    {
     "slot": "E1",
     "subject_en": "receding old truck on the left branch",
     "x": 0.55,
     "y": 0.35,
     "anchor_en": "on the curved section of the left road branch just past the junction",
     "facing": {
      "x": 0.8,
      "y": 0.3
     },
     "facing_anchor_en": "further down the straight path of the left road branch",
     "camera_facing_relation": "away_from_camera",
     "camera_facing_evidence": "뒷모습"
    },
    {
     "slot": "E2",
     "subject_en": "receding old truck on the right branch",
     "x": 0.6,
     "y": 0.67,
     "anchor_en": "on the straight section of the right road branch moving away from the fork",
     "facing": {
      "x": 0.85,
      "y": 0.7
     },
     "facing_anchor_en": "further down the right road branch toward the eastern edge of the map",
     "camera_facing_relation": "away_from_camera",
     "camera_facing_evidence": "뒷모습"
    },
    {
     "slot": "E3",
     "subject_en": "following vehicles",
     "x": 0.52,
     "y": 0.6,
     "anchor_en": "on the curved entrance of the right road branch just below the fork",
     "facing": {
      "x": 0.75,
      "y": 0.68
     },
     "facing_anchor_en": "further down the straight section of the right road branch",
     "camera_facing_relation": "unspecified",
     "camera_facing_evidence": ""
    }
   ],
   "camera": {
    "origin": {
     "x": 0.25,
     "y": 0.5
    },
    "look_target": {
     "x": 0.6,
     "y": 0.5
    },
    "origin_anchor_en": "on the common road segment, well before the two-branch fork",
    "look_target_anchor_en": "in the green space separating the left and right road branches",
    "view_left": {
     "x": 0.65,
     "y": 0.2
    },
    "view_right": {
     "x": 0.65,
     "y": 0.8
    },
    "view_left_anchor_en": "along the upper grassy boundary above the left road branch",
    "view_right_anchor_en": "along the lower grassy boundary below the right road branch"
   },
   "rationale_ko": "카메라는 갈림길 이전의 공통 도로에 위치하여, 양쪽 도로로 갈라져 멀어지는 두 낡은 트럭의 뒷모습 풀샷을 포착합니다. 한 대의 트럭은 왼쪽 도로 가지로 향하게 하고, 다른 트럭과 그 뒤를 따르는 차량들은 오른쪽 도로 가지에 배치하여 서로 다른 방향으로 분기하는 동선을 강조했습니다."
  },
  "marker_check": {
   "markers_ok": false,
   "details_ko": "엔티티 마커들의 위치와 식별자가 서로 뒤바뀌어 배치되었습니다.",
   "violations": [
    {
     "axis": "entity_position",
     "note_ko": "E1, E2, E3 마커가 지정된 좌표가 아닌 엉뚱한 위치에 배치되었으며, 식별자(라벨)가 서로 뒤바뀌었습니다."
    },
    {
     "axis": "entity_facing",
     "note_ko": "마커 슬롯이 뒤바뀜에 따라 각 엔티티가 바라보는 방향(화살표)도 지정된 방향과 일치하지 않습니다."
    }
   ]
  },
  "marker_prompt": "Edit the attached TOP-DOWN SITE PLAN image: draw STAGING MARKERS on it\nat the positions described below. This is a planning aid for a film\ncrew, drawn over the plan.\n\nPLACEMENT AUTHORITY: each marker line below gives its position as\nnormalized fractions of this canvas — x runs left→right 0..1, y runs\ntop→bottom 0..1 — and THOSE FRACTIONS ARE THE AUTHORITY. Measure them\non the attached image and put the marker there. Each line also names a\nnearby landmark; use it to check you landed on the right feature, and\nwhen a landmark phrase and the fractions disagree, follow the\nfractions. This applies to compass words (north, south, east, west)\ntoo: read them only if this plan actually shows a compass or labels\nits own approaches, and even then they never move a marker away from\nits fractions. Do NOT relocate a marker onto a different salient line\nor feature than the one its fractions land on.\n\nNEVER move a marker to make the drawing fit. The camera circle and\nevery subject circle go exactly where their descriptions place them,\neven when that leaves the view wedge or an arrow running off the edge\nof the image — in that case just let the wedge or arrow continue to\nthe image border and stop there. Shifting the camera outward, pulling\na subject aside, or shrinking the wedge to keep it inside the canvas\nall break the plan this map exists to record.\n\nDraw EXACTLY these markers, flat and high-contrast, small enough not\nto hide the plan under them:\n- CAMERA: one filled circle at the camera position, labelled CAM, and\n  out of that circle a FIELD-OF-VIEW WEDGE showing what this shot\n  sees — two straight thin lines, one running toward each of the two\n  frame-edge positions given below, and a third thinner line to the\n  look-target position between them. Fill the area between the two\n  edge lines with a light translucent tint so the wedge reads as the\n  camera's view; the plan underneath must stay fully readable through\n  the tint. Every subject listed below stands inside this wedge —\n  if your drawing leaves one outside it, the wedge is drawn wrong,\n  not the subject.\n- SUBJECTS: one filled circle per subject at its position, labelled\n  with its slot code (E1, E2, ...) followed by its short description\n  in parentheses, and one short ARROW leaving that circle toward the\n  spot the subject faces, so the direction each subject is turned is\n  readable at a glance.\n\nChange NOTHING else: keep every zone, landmark, line, label and color\nof the plan exactly as attached — do not redraw, recolor, move or\nremove anything, do not add scenery, shadows or any other symbols or\ntext beyond the markers, wedge, arrows and labels specified above. The\noutput is the SAME plan image with only these staging markers added.\n\nMARKERS TO DRAW:\n- CAM (camera): circle at x=0.25, y=0.50 — the left edge area of the plan (landmark, for reference: on the common road segment, well before the two-branch fork); its sight line ends at LOOK TARGET: at x=0.60, y=0.50 — the center of the plan (landmark, for reference: in the green space separating the left and right road branches); the frame's LEFT edge cuts through at x=0.65, y=0.20 — the upper middle of the plan (landmark, for reference: along the upper grassy boundary above the left road branch) and its RIGHT edge through at x=0.65, y=0.80 — the lower middle of the plan (landmark, for reference: along the lower grassy boundary below the right road branch)\n- E1 (receding old truck on the left branch): circle at x=0.55, y=0.35 — the center of the plan (landmark, for reference: on the curved section of the left road branch just past the junction), facing at x=0.80, y=0.30 — the upper-right area of the plan (landmark, for reference: further down the straight path of the left road branch)\n- E2 (receding old truck on the right branch): circle at x=0.60, y=0.67 — the lower middle of the plan (landmark, for reference: on the straight section of the right road branch moving away from the fork), facing at x=0.85, y=0.70 — the lower-right area of the plan (landmark, for reference: further down the right road branch toward the eastern edge of the map)\n- E3 (following vehicles): circle at x=0.52, y=0.60 — the center of the plan (landmark, for reference: on the curved entrance of the right road branch just below the fork), facing at x=0.75, y=0.68 — the lower-right area of the plan (landmark, for reference: further down the straight section of the right road branch)",
  "marker_attempts": 3,
  "marker_best_effort": true,
  "marker_pick_index": 1,
  "marker_pick_count": 3,
  "marker_pick_why_ko": "후보 1은 카메라 마커의 위치와 시야각(웨지)이 모든 피사체를 올바르게 포함하고 원본 도면을 온전히 보존하여 가장 원본에 가깝습니다. 피사체 마커(E1, E2, E3)들의 위치가 서로 뒤바뀌어 배치되고 그로 인해 방향이 어긋난 점만 제외하면 지시사항을 잘 따랐습니다. 반면 다른 후보들은 시야각이 피사체를 놓치거나 카메라 위치가 잘못되는 등 더 많은 오류를 포함하고 있습니다.",
  "marker_model": "nb2",
  "marker_map_sha256": "b89d211c3da3581c695c928c6058c2aa914ef5cd6b6f736a2b9b3bf2ca730407",
  "leakage": {
   "has_marker_leakage": false,
   "details_ko": "도식화된 기호나 텍스트, 평면도 누출 없이 입체적인 뷰로 잘 표현되었습니다."
  },
  "prompt": "Draw a single CAMERA-VIEW storyboard frame (16:9) of the scene staged\non the attached MARKER-ANNOTATED TOP-DOWN SITE PLAN, seen from the\ncamera position marked on it.\n\nThe attached plan carries STAGING MARKERS drawn over it: a\ncamera-position marker with a field-of-view wedge (where the camera\nstands, which way it looks and how wide it sees) and one labelled\ncircle per subject with an arrow (where each subject stands and which\nway that subject is turned). Jurisdiction of sources — follow all three, they do not\nconflict:\n- The SITE PLAN under the markers is the SOT for what exists on the\n  site and how its zones and landmarks are arranged relative to each\n  other. The markers themselves are staging aids, NOT site content.\n- The markers on the plan are the SOT for XY POSITIONS, horizontal\n  heading and body facing ONLY: where each subject stands on that\n  plan, which way each one is turned, where the camera stands, which\n  way it looks and how wide its wedge opens. Distances, directions\n  and relative depths between subjects come from the marked\n  positions; what the wedge covers is what this frame contains, and\n  subjects outside the wedge are out of frame. The STAGING GEOMETRY\n  text below restates the same staging with coarse normalized\n  estimates — it never overrides the marked plan.\n- The CAMERA DIRECTION text below (when present) is the SOT for the\n  VERTICAL angle, elevation, tilt and framing (eye-level, high angle,\n  overhead crane, low angle, ...). If it asks for a high or overhead\n  angle, draw that angle — the site plan being top-down does NOT mean\n  the frame must be eye-level. It does NOT govern horizontal\n  placement: which side of the frame a subject occupies and which way\n  that subject is turned come from the marked plan and the staging\n  text, even when that memo also mentions left, right, foreground or\n  background.\n\nSTYLE — MANNEQUIN BLOCKING SKETCH: the lightest possible director's\nlayout drawing. THIN, CLEAN, UNIFORM PENCIL CONTOUR LINES ONLY on pure\nwhite paper. Absolutely NO shading, NO tone, NO texture, NO grayscale\nfill, NO light-and-shadow, NO material rendering. It must read as a\nquick layout guide, never as a finished drawing.\n\nDraw every person as a PLAIN ARTIST'S MANNEQUIN: a featureless oval head\nwith NO face, NO hair, NO clothing and no costume detail — simple tapered\nlimbs and torso. Because the head has no face and no hair, the ONLY thing\nthat shows which way a figure is turned is the BODY ITSELF: the direction\nthe chest, hips and feet point, and the way the head-oval sits on the\nshoulders. Aim each mannequin's chest and feet at the spot the staging\ntext says that person faces, and let the arms carry the action. A\nmannequin turned to the wrong side is the single worst error in this\ndrawing. Environment is reduced to a few placement lines (ground line,\nkey landmark silhouettes) — just enough to read positions and depth.\n\nSTAGING GEOMETRY — PLANNING DATA ONLY, NEVER RENDER IT: the lines\nbelow restate, marker by marker, the staging already marked on the\nattached plan: each landmark-relative description restates the\nmarker-map authority in words, and the normalized fractions in\nparentheses (x left→right 0..1, y top→bottom 0..1) are coarse audit\nestimates only — neither ever overrides the marked plan.\n\nThose fractions are TOP-DOWN bookkeeping. Never read them as left and\nright of your frame: on a plan seen from above, a subject \"to the\nright\" can sit on either side of a camera that looks any direction.\nEach subject line therefore ends with the SAME staging expressed from\nTHIS camera — which part of the frame the subject occupies, and which\nway that subject faces in screen terms (toward the left or right of\nthe frame, toward or away from the camera or side-on to it in\nprofile, and whom they are looking at). Those screen-space statements\nare what your drawing must match:\na figure said to face the right of the frame is turned to the right in\nyour drawing, never mirrored.\n\nDo NOT draw any of this data in the image — no numbers, no\ncoordinates, no labels, no arrows, no cones, no dots, no diagram\nsymbols of any kind.\n\nWhere the CAMERA DIRECTION text below also describes horizontal\nplacement (which side of the frame someone occupies, foreground left,\nbackground right), treat that as background reading only: for anything\nabout left, right and where a subject sits in this frame, the staging\nabove wins. CAMERA DIRECTION governs the vertical angle, elevation,\ntilt and framing.\n\nFor whether a subject faces the camera or is turned away from it, each\nsubject line above already states it in screen terms and the SHOT TEXT\nis the source it was derived from — draw exactly what the subject line\nsays, and never mirror it.\n- the camera stands on the common road segment, well before the two-branch fork (coarse estimate x=0.25, y=0.50); its sight line ends at LOOK TARGET: in the green space separating the left and right road branches (coarse estimate x=0.60, y=0.50); the frame's LEFT edge cuts through along the upper grassy boundary above the left road branch (coarse estimate x=0.65, y=0.20) and its RIGHT edge through along the lower grassy boundary below the right road branch (coarse estimate x=0.65, y=0.80) — everything between those two edges is in frame, anything outside them is out of frame\n- receding old truck on the left branch stands on the curved section of the left road branch just past the junction (coarse estimate x=0.55, y=0.35), facing further down the straight path of the left road branch (coarse estimate x=0.80, y=0.30) — from this camera the subject is in the LEFT part of the frame and faces away from the camera, into the depth of the shot, so the camera sees this subject from behind\n- receding old truck on the right branch stands on the straight section of the right road branch moving away from the fork (coarse estimate x=0.60, y=0.67), facing further down the right road branch toward the eastern edge of the map (coarse estimate x=0.85, y=0.70) — from this camera the subject is in the RIGHT part of the frame and faces away from the camera, into the depth of the shot, so the camera sees this subject from behind\n- following vehicles stands on the curved entrance of the right road branch just below the fork (coarse estimate x=0.52, y=0.60), facing further down the straight section of the right road branch (coarse estimate x=0.75, y=0.68) — from this camera the subject is in the RIGHT part of the frame and faces toward the RIGHT of the frame and away from the camera, into the depth of the shot, so the camera sees this subject from behind\n\nCAMERA DIRECTION (vertical angle / elevation / framing SOT): From the elevated position behind the fork, hold the crane's stable middle composition with a downward angle that includes both trucks in complete rear three-quarter view. 현우's truck recedes along the left branch and 수빈's along the right, each occupying less than a quarter of the image, with the following vehicles taking the right branch and wheel-raised dust tracing the divergence. Keep the fork visible in the lower center and the occupants outside readable view, reserving the later descent and pursuit for 현우's branch.\n\nTHE LOCATION: 익산과 서남부 내륙의 지방도로\n\nSHOT TEXT (authoritative, Korean): 양 갈래 흙길에서 서로 다른 방향을 향해 달려가며 바퀴 뒤로 흙먼지를 일으키고 있는 두 낡은 트럭의 뒷모습 풀샷.\n\nCARRIED STATE (persist exactly): Two old roofless trucks separate at the fork, with more vehicles following one branch. Charlie rides in the cargo bed with his battle damage unrepaired and the recovered B-200 chest component in his possession.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract or shot-text direction says otherwise,\nsketched figures read as alive in mid-moment — believable weight\nshift, hands engaged, head-oval and chest turned toward the target the\nSHOT TEXT implies; avoid a stiff attention stance (feet together, arms\nhanging straight down) and a head-oval squared flat to the lens. If the\nSHOT TEXT or any POSE contract above stages stillness, death, sleep,\nunconsciousness, restraint, drill or an explicit direct-to-camera look,\ndraw THAT exactly — this clause never overrides it.\n\nCRITICAL — THE SITE PLAN IS A STAGING REFERENCE, NOT CONTENT: do NOT\ndraw the plan itself, and do NOT draw any diagram elements in the\nframe — no circles, dots, wedges, cones, arrows, callouts, codes or\nany other overlay symbols. The output is a clean storyboard drawing\nof the scene as seen by the camera only.\n\nNo text, no letters, no numbers, no captions anywhere in the image."
 }
}