完整提示词
a single raindrop slides slowly down a foggy window at night, camera holds completely static, neon city lights blurred in the background, slow motion 120fps look, single vertical motion only, shallow depth of field, cool blue and amber color grading, no camera shake, 5-second cinematic clip
结构拆解
这条提示词稳定的核心是静帧+单一动作。AI 视频在静止镜头+单一缓慢动作上成功率最高（大多数模型 70%+），而一旦镜头和主体同时大幅运动，成功率会跌到 20% 以下。
3 种改写版本
a young woman in a black trench coat walks slowly forward through a rain-soaked Tokyo alley at night, camera tracks horizontally to the right at the same walking pace, neon reflections on wet ground, shallow depth of field, steady gimbal feel, 6-second cinematic shot, no jitter
从静帧改成跟拍：主体走+镜头同步跟。务必写 "at the same walking pace" 让镜头不超前/落后。
misty mountain valley at sunrise, slow steady drone dolly forward over the treetops, sunlight breaking through low clouds, very gentle pacing over 8 seconds, cinematic wide shot, smooth motion, no jitter
推镜核心是"slow steady"+明确时长（over 8 seconds）+ "smooth motion"。不写时长，模型推得过快。
extreme close-up of melted chocolate slowly dripping onto a glossy croissant, camera holds completely static, single dripping motion only, warm side light from the right, shallow depth of field, slow motion 120fps look, 3-second clip, no shake
食品视频最适合静帧+单一缓慢动作。"single dripping motion only"防止模型加搅拌/晃动等多余动作。
常见翻车点
"moving"、"interacting"是模糊词。改成具体动词："slides slowly"、"pours"、"turns to the left"、"lifts"。
"walks in, sits down, picks up the cup" 在 5 秒里塞 3 个动作必崩。一条镜头只允许 1 个核心动作。
不写 "3-second / 5-second clip"，模型按自己默认节奏来，往往太快。
不写 "no camera shake / steady gimbal"，多数模型默认轻微手持抖动，把静帧氛围毁掉。
主体快速运动+镜头大幅运动同时发生，模型几乎必崩。先固定一个，让另一个动。
"masterpiece, 8k, best quality"对视频几乎无效。视频写"cinematic, shallow depth of field, color grading"就够。
模型对比表
| 模型 | 典型时长 | 擅长 | 这条案例评分 |
|---|---|---|---|
| Runway Gen-3 | 5-10s | 电影感、跟拍 | ★★★★★ 静帧极稳 |
| Pika 2.x | 3-5s | 短氛围片 | ★★★★ 适合 3-5s 静帧 |
| Kling 2.x | 5-10s | 人物、广告 | ★★★★ 中文友好 |
| Seedance 2.0 | 5-8s | 横屏电影感 | ★★★★★ 横屏推荐 |
| Hailuo / MiniMax | 5-6s | 真人、风景 | ★★★ 中文长句友好 |
适用场景
这条静帧骨架适合：开场氛围镜头、产品广告中段、短视频 b-roll、社交媒体故事片头、电影预告片转场。重要原则：静帧+单一缓慢动作是 AI 视频成功率最高的组合，多镜头叙事建议剪辑多段静帧拼合，而不是用一条提示词强行让模型自己变出复杂动作。