integrated_multimodal_description: [Shot 1] Live-action, photorealistic handheld smartphone video, filmed as one continuous uncut take entirely from the one recording smartphone's lens. A casually dressed man in his early 30s with a natural conversational voice (S1) walks along a busy modern downtown sidewalk in daylight while recording himself with the unseen front-facing smartphone camera held at arm's length in selfie orientation. The recording smartphone itself remains outside its own image at all times. The framing is imperfect and authentic, with subtle handheld shake, natural walking bounce, realistic smartphone autofocus, exposure adjustment, and slight motion blur. Pedestrians and moving traffic fill the city street behind him. During the opening several seconds, he looks directly into the camera lens while continuing to walk and casually says: <d>[English] This is a test of different attention backends.</d> Immediately after finishing the sentence, he notices something happening ahead of him. His eyes suddenly widen, his eyebrows rise, and his relaxed expression changes sharply to genuine shock. He stops walking for a moment and quickly rotates the unseen recording smartphone away from his face toward the street ahead. The recorded image pans right with large amplitude at fast speed in one continuous handheld movement, producing realistic motion blur before autofocus rapidly settles on the street. The viewer continues to see exclusively what the recording smartphone lens sees: there is no outside camera, no third-person angle, no view of the recording smartphone, and no second phone visible anywhere in frame. Directly in front of him at a downtown intersection, a moving sedan suddenly collides hard with another car crossing its path. The first vehicle's front end crumples on impact and both vehicles jolt realistically from the collision, scattering small pieces of plastic and glass across the pavement. Tires skid and one car rotates partially sideways before stopping. Nearby pedestrians visibly react to the crash: several jump backward, two people run away from the street, another person raises their arms in surprise, and bystanders turn toward the accident. The event is chaotic but non-graphic; nobody is visibly injured and there is no blood. After holding the crash in frame briefly, he quickly rotates the same unseen recording smartphone back toward himself while the footage remains exclusively the continuous first-person lens view. The recorded image pans left with large amplitude at fast speed, briefly blurring the city before reacquiring his face. The recording smartphone and the hand holding it remain outside the image. His identity, clothing, hairstyle, and facial features remain completely consistent with the opening. His shocked expression fades naturally; his breathing settles and his face returns to an oddly calm, matter-of-fact expression. He stops walking, looks directly into the camera lens, pauses for a beat, and says with casual curiosity: <d>[English] Well, how was the quality?</d> He holds eye contact with the camera through the final moment. overall_soundscape: Continuous realistic downtown ambience with distant traffic, footsteps, pedestrians talking, and light wind around the smartphone microphone. The collision produces sudden tire squeal, a loud metallic impact, crunching bodywork, breaking glass, and brief car horns, followed by startled gasps and shouts from nearby pedestrians. The camera movement creates subtle handling noise and shifting wind noise while the environmental audio remains spatially consistent throughout the continuous take. camera_continuity_constraint: The complete clip is the image captured by one unseen smartphone camera. Never show a phone, another recording device, the hand holding the camera, an outside view of the cameraman, a third-person camera angle, a mirror view, or a reflection of the recording device. non_diegetic_music: N/A