One runner leaves a trail of fire, the other a trail of ice, four 15-second segments chained into 60: from a neon city through a tunnel, along a rocky coast, and out onto open water. The prompt only sketches the workflow skeleton (GPT Image 2 for stills -> Seedance to animate) - the visuals are left for the model to fill in.
one runs hot. on runs cold.
a Friendly race
GPT2 images -> Seedance
Four 15-second scenes chaining into a 60-second short. 16:9 landscape, photoreal, four terrains, one continuous piece of motion
한 명은 뜨겁게 달리고, 한 명은 차갑게 달린다.
우호적인 경주
GPT2 이미지 -> Seedance
네 개의 15초 장면이 이어져 60초짜리 단편이 됨, 16:9 가로 화면, 포토리얼, 네 가지 지형, 끊김 없이 이어지는 하나의 연속 동작.
✍️ Editor’s Notes
The author only gives you a skeleton here: two character designs (one leaves a trail of fire, one leaves a trail of frost), one line about it being a friendly race, one workflow note (GPT Image 2 generates stills, Seedance takes over for continuation), plus a spec line — four 15-second segments stitched into 60 seconds, 16:9, photorealistic, four terrains, motion never stops. There's not a single shot description anywhere in the whole thing — the neon city, the tunnel, the coastal rocks, the ocean surface are all things the model filled in on its own. The interesting part isn't the wording itself, it's what this demonstrates about breaking a long video apart: use a fixed character design plus “one continuous motion” as the seam glue, and chain four independently generated segments together — that holds consistency better than trying to hand-write one 60-second mega-prompt. The tradeoff is you've basically surrendered all control; if you want specific shot types or lighting, you have to go fill that into the skeleton yourself.
Want tighter control? Try this expanded version:
Two runners, consistent identities throughout: Runner A trails live embers and orange heat-shimmer; Runner B trails frost crystals and pale blue vapor. Same athletic build, same fitted running suits (A in charcoal with ember seams, B in white with frost seams), same faces in every scene. A friendly race, not a fight — both smiling, pushing each other.
16:9 landscape, photorealistic, cinematic anamorphic look, natural motion blur, one continuous piece of motion across all four scenes: each scene ends mid-stride and the next begins mid-stride at the same pace and screen direction (left to right throughout).
Scene 1 (0-15s) — Neon megacity at night: low tracking shot alongside the runners through a canyon of holographic signage, wet asphalt reflecting their fire and ice trails, drones streaking past overhead.
Scene 2 (15-30s) — Highway tunnel: rear tracking shot, sodium lights strobing over their backs, the two trails braiding into a single ribbon on the tunnel wall behind them.
Scene 3 (30-45s) — Storm-battered coastal rock shelf: wide side tracking shot at dusk, sea spray bursting where the ember trail meets the water, frost blooming across the wet stone under Runner B.
Scene 4 (45-60s) — Open ocean surface: they run on the water itself, camera pulling back and rising into a high wide aerial, fire wake and ice wake carving two long parallel lines across the sea toward the horizon.
Audio: footfalls, wind, surf, the hiss of steam where hot meets cold. No music.
