| "global_caption": "subject_definitions:\n<Subject 1> is the indoor hotel lobby in <Picture 1>: a real-world location under indoors, evening ambiance, with its specific lighting, colours and materials (polished marble floor tiles, dark polished wood, woven wicker, gilded picture frame, fabric upholstery, glazed ceramic planter, fire and embers), and its overall spatial layout. Resting in this location: the burning wicker basket on the console table, still on fire with steady orange flames.\n<Video 1> is a schematic 3D annotation of the same place: a smooth colour-gradient backdrop encodes the camera's heading (its hue shifts as the camera turns), a cyan wireframe lattice of cubes marks 3D space, yellow-green grid lines mark the walkable ground, and together they encode the camera path through the scene. It carries no appearance information. Solid coloured blocks mark where the marked objects are and how big they are: people, vehicles, or the target object that will be acted on. A block's colour encodes which of its states the object is in: purple means its ordinary default state (a person or vehicle simply present, or a target object as it is normally found), yellow means the first non-default state the object takes on in this clip, and blue means a second, later non-default state. An object may already be in a non-default state when the clip begins, in which case its block is yellow from the start. The colours mark only the order of the states, not what the object looks like - what each state actually is is stated in the text of the time segment in which the change happens. A block moves with its object, a gradual colour transition marks the moment a change is under way, and a block may disappear when its object leaves the scene, is taken into the hand, or stops being a single object at all because it has been broken apart. The blocks are positional markers, not objects to be drawn.\n<Picture 2> and <Picture 3> are earlier observations of the same place as <Subject 1>, recorded shortly before the target video from nearby viewpoints. They establish the layout of the place and the identity, appearance and placement of the objects in it; what happens to any object in the target video is given by <Video 1> and the text, not by these pictures.\n\nsummary:\n[reference generation] The target video is photorealistic live-action footage of <Subject 1>, with the camera moving through the scene along the trajectory encoded by <Video 1>, and the burning wicker basket on the console table, still on fire with steady orange flames at the positions marked in <Video 1>.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - the photographic look, lighting, colours, materials and spatial layout of the location are retained throughout.\n<Video 1> (camera movement and object placement structure): weak_reference - only the camera trajectory and the marked positions and states of the objects are followed. No wireframe lattice, no colour-gradient backdrop, no grid lines and no flat-shaded blocks appear anywhere in the target video. Each coloured block is replaced by the corresponding photorealistic object in the state its colour indicates, at the same position and occupying the same fraction of the frame, consistent with <Subject 1>.\n<Picture 2> and <Picture 3> (memory of the place): fully_preserved - the layout of the place and the identity, appearance and placement of the objects seen in them are kept consistent whenever they come into view.\n\ndetailed_description:\nThe target video is photorealistic live-action footage. Warm ambient lighting from ceiling chandeliers and recessed fixtures, combined with a vibrant, flickering orange glow cast by the active fire onto the tabletop and adjacent wall.\n[Shot 1] A spacious, elegant hotel lobby features reflective cream-colored marble tile flooring stretching down a central corridor. Along the right wall, beneath a framed landscape painting with a gilded border, sits a dark wood console table with cabriole legs. Positioned on top of the console table, a woven wicker basket with an arched handle is on fire, containing steady, bright orange flames that lick upward through the handle and cast a warm radiant glow across the wooden tabletop and textured beige wall. Further down the corridor stands a tall potted palm tree in a dark round ceramic planter set beside a wood-trimmed structural pillar. To the left, a carpeted seating lounge is furnished with patterned armchairs beneath a coffered ceiling with a crystal chandelier, while the dark wood front desk is visible in the background. The wicker basket on the console table is on fire, burning steadily with bright orange flames, and it keeps burning. This is one continuous first-person shot, without cuts or dissolves. The initial view faces away from the earlier interaction station. The observer turns back along the camera path in <Video 1>, returns to that same station, then remains there observing until the final frame. The state shown in the historical photographs already exists before this clip begins and remains constant, including the very first partial reappearance. No transformation or interaction is replayed. Only the camera moves; hands, tools and transient action effects do not appear.\n\noverall_soundscape:\nN/A - the target video has no soundtrack.\n\nnon_diegetic_music:\nN/A", |
