This prompt is written directly in JSON structure: subject, scene, and camera are described as separate fields for the rider crossing the city. In the ongoing debate between structured prompts and natural-language prompts, this is a sample from the structured camp — a 34-second finished clip.
"video_generation": {
"subject": {
"character": "Young urban man, high flat-top fade, goatee, neck and arm tattoos, silver hoop earrings, geometric reflective sunglasses",
"clothing": "White and beige varsity jacket with '38' on sleeves, black shorts, white socks",
"vehicle": "Matte black fixed-gear bicycle"
},
"scenes": [
"Close-up of a hand holding a smartphone displaying a glowing red heart.",
"Character sips a pink iced drink, smirks, and gets on his bicycle.",
"Aggressive, fast-paced cycling through sunlit city streets, weaving through traffic.",
"Character rides quickly down a flight of concrete stairs.",
"High-speed navigation through a crowded Asian street food market, bumping a man and causing noodles to fly in slow motion.",
"Riding up stairs and launching off a dirt jump ramp with a city skyline in the background.",
"Character skids to a perfect stop in front of a modern glass building where a woman is waiting."
],
"camera_work": "Dynamic tracking shots, fast-paced action, brief slow-motion emphasis during the food spill",
"style": "Ultra-realistic 3D animation, cinematic lighting, vibrant colors, motion blur",
"render_settings": {
"resolution": "8K",
"quality": "Cinematic, high-detail",
"aesthetic": "Hollywood movie"
}
}
}
"video_generation": {
"subject": {
"character": "젊은 도시 남성, 하이 플랫탑 페이드 헤어, 염소수염, 목과 팔에 타투, 실버 후프 귀걸이, 기하학적 형태의 반사 선글라스",
"clothing": "화이트-베이지 배색 바시티 재킷, 소매에 '38' 문구, 검정색 반바지, 흰색 양말",
"vehicle": "무광 검정색 픽시(고정기어) 자전거"
},
"scenes": [
"클로즈업 샷: 손 하나가 스마트폰을 들고 있고, 화면에는 빛나는 빨간 하트가 표시됨.",
"캐릭터가 분홍색 아이스드링크를 마시며 득의양양한 미소를 짓고, 이어 자전거에 올라탐.",
"햇살 쏟아지는 도심 거리를 빠르게 질주, 동작이 날카로움, 차량 사이를 누빔.",
"캐릭터가 빠르게 자전거를 몰아 콘크리트 계단을 내려감.",
"붐비는 아시아 거리 음식 시장을 고속으로 질주하다 한 남성과 부딪혀, 면발이 슬로모션으로 흩날림.",
"계단을 올라 흙 언덕 점프대에서 도약, 배경에는 도시 스카이라인.",
"캐릭터가 현대적인 유리 건물 앞에서 완벽하게 급정지, 한 여성이 그곳에서 기다리고 있음."
],
"camera_work": "다이내믹 트래킹 샷, 빠른 템포의 액션 장면, 음식이 흩날리는 순간에는 짧은 슬로모션으로 강조",
"style": "초사실적 3D 애니메이션 스타일, 시네마틱 라이팅, 선명한 컬러, 모션 블러",
"render_settings": {
"resolution": "8K",
"quality": "시네마틱, 고디테일",
"aesthetic": "할리우드 영화 질감"
}
}
}
✍️ Editor’s Notes
The most notable thing about this one is its writing format itself — it's not natural-language description at all, but a raw JSON fragment (subject/clothing/vehicle/scenes/camera_work/style/render_settings), and the opening is even missing the outer key name and left curly brace, clearly sliced out of a larger structured data blob. That tells you the author is driving the model in something close to an API-call style, filling in character setup, scene list, camera language, and render parameters as separate fields rather than folding them into prose. The seven scene entries themselves are extremely cinematic — from a heart-shaped notification, to skydiving off steps, to crashing into a night market and sending noodles flying — action beats numbered in shot order, structurally the same as the storyboard-style writing seen in the other long Chinese-language prompts, just wearing a shell more familiar to programmers: key-value pairs. This suggests the GPT Image 2 + Seedance 2.5 combo may support both structured JSON input and natural-language input as parallel parsing paths.
