Frozen-Tone Large-Format Snowscape · Cross-Model Test

Seedance 2.5RealisticWorkflowDrone AerialSlow MotionIce & SnowUnderwater WorldSamurai & Ninja20s
The author's thesis is stated right up front: "whatever Seedance 2.5 can do, MiniMax H3 can do too." The same icy-blue monochrome large-format prompt is reproduced across models — hard proof of prompt portability.
PROMPT · 视频提示词
Create with AI
What Seedance 2.5 can do, MiniMax H3 can do just as well — a Seedance prompt can achieve the same design intent when used in MiniMax H3. Photorealistic live-action film style, shot on large-format anamorphic lenses, heavy volumetric fog with snowflakes drifting slowly down, an almost monochromatic frozen color palette dominated by pale ice-blue, bone white, and wet slate grey, with virtually no saturation anywhere except for a single molten-gold accent color and nothing else, flat overcast lighting with no sun and no hard shadows, an extreme scale contrast between the figure and the structure, shallow depth of field in close-ups, deep aerial perspective, distant layers fading to white, fine natural film grain, a big-budget sci-fi epic, restrained color grading, no punchy impact. Fully original character and creature design — must not resemble any existing film, game, or animated work. 16:9. 20 seconds, SEVEN shots, slow and deliberate. Action begins on frame 1, no static opening. No one speaks, no lip movement, no dialogue. Character lock — identical in every shot: a solitary figure of indeterminate age, gaunt and still in build, wrapped in a heavy, floor-length, thick bone-white felt coat with a dark fur-lined hood pulled up. Frost has formed on the shoulders and hem. Grey-white hair, damp and pushed back from the face. Pale skin, frost on the eyelashes, an expression calm and unreadable. Beneath the coat are plain, dark layered garments and worn leather gloves. No armor, no weapons, no emblems, no forehead markings, no jewelry, no facial piercings. The only color on this character is the molten-gold glow in the eyes, which do not open until the fourth shot. Colossus lock — identical in every shot: a massive kneeling stone statue, half-swallowed by ice, colossal in scale, with only its bowed head, one shoulder, and one upturned open palm rising above the frost line. Weathered granite, deep erosion grooves, moss and ice packed into every crevice. Its face has been worn nearly smooth over centuries — no expression, no teeth, no horns, no demonic features, just a serene, ruined face with closed eyes. Fine veins of dark gold run through the cracks in the stone, dormant. World lock: a submerged city entirely frozen over — ruined towers, broken bridges, industrial pipes, and heavy cables stretching between structures, all encased in ice and fading into white fog within a hundred meters. No human figures anywhere. No text, no signage. Shot 1 — 0.0s to 3.0s — extreme wide angle, long lens, camera drifting extremely slowly: the ice-bound ruined city shrouded in fog, at its center the colossus rising from the ice with its bowed head and one massive open stone palm. Snow falls steadily. The scale is overwhelming, the frame almost entirely grey-white. CUT. Shot 2 — 3.0s to 6.0s — wide angle, an extremely slow push-in: a tiny figure in a bone-white coat stands motionless at the center of the colossus's open stone palm, no larger than one of its fingernails. Fog roils between them. CUT. Shot 3 — 6.0s to 9.0s — close-up portrait, static, shallow depth of field: the figure's face inside the fur hood, eyes closed, frost gathered on the eyelashes, breath fogging and drifting away. Absolutely still. Everything behind them is soft white void. CUT. Shot 4 — 9.0s to 12.0s — extreme close-up, static: the eyes open. They burn with a deep, molten-gold glow — the first and only color in the entire film — casting a faint warm glow onto the frost on the skin. Hold here. CUT. Shot 5 — 12.0s to 15.0s — wide low angle, camera rising slowly: the figure now stands on a rough dark stone slab, silently suspended high above the frozen city, the coat billowing sideways in the wind, the colossus's bowed head looming faintly in the fog behind. CUT. Shot 6 — 15.0s to 17.5s — medium shot, static: the figure slowly and deliberately raises one open, gloved hand, reaching down toward the distant city. No drama. Just the gesture. CUT. Shot 7 — 17.5s to 20.0s — extreme wide angle, camera locked off: fine veins of molten gold race outward through the ice and upward through the cracks in the colossus, and throughout the city the ice-bound towers, bridges, and cables detach from the ground and rise silently into the air, suspended in the fog. The frame holds on the floating city as snow continues to fall. The seven cuts fall at 3.0s, 6.0s, 9.0s, 12.0s, 15.0s, and 17.5s. No dissolves, no fades, no speed ramping. The camera moves slowly and in a controlled manner throughout. Every shot sustains subtle continuous motion — falling snow, drifting fog, breathing, billowing fabric — nothing is ever completely still. Sound: a low, sustained wind sweeping across the open ice, the deep groan of ice under pressure, the soft tick of snowflakes striking fabric, a long, low sub-bass swell as the eyes open, a rising low resonant hum as the gold spreads, and finally a deep grinding sound as an immense mass detaches from the ground. No voice, no dialogue, no narration, no music. Negative prompt: anime, Japanese anime style, cartoon rendering or stylized animation, ninjas, headbands, forehead protectors, cloaks with printed patterns or cloud motifs, ringed or patterned irises, orange or spiked hair, facial piercings, any resemblance to existing film, game, comic, or animated characters, weapons, gore, entrails, wounds, visible crowds or bystanders, warm sunlight, blue sky, any saturated color other than gold, gradients, lens flares, on-screen text, subtitles, letters, words, numbers, readable signage, logos, watermarks, cartoon look, 3D rendering, video-game cinematic style, slow motion, speed ramping, static opening, static first frame.
✍️ Editor’s Notes

开篇第一句就挑明这是一条「跨模型验证」实验——原本是给Seedance 2.5写的词,作者特意拿去MiniMax H3上测试设计意图是否能被同等还原,这种写法本身就是在给提示词工程做A/B测试,而不是单纯追求成片效果。核心的角色锁定和巨像锁定分别用大段文字重复描述「每场镜头完全相同」,这是应对多镜头(七个)拆分生成时最容易出现的角色漂移问题,靠反复而不是简写去对抗生成的不确定性。整条词只留一个熔金色作为全片唯一有彩色的元素,其余全部锁进冰蓝骨白的低饱和体系,这种「单一强调色」策略比常见的色彩分级描述更容易被模型准确执行,因为它把选择空间压缩到了极致。负面清单里专门点名忍者、头带、额头护具、云纹斗篷这类具体的动漫符号,说明原创角色设计最大的风险就是模型会不自觉往熟悉的IP形象上靠,这条词是在正面反击这种路径依赖。

Create with AI
Related Works
×