AI 视频创作者 Matthew Whitmore 用实测结果说明了一个让不少创作者难以接受的事实:让 Runway 输出平庸内容的,不是模型的上限,是提示词的写法。
大多数人把提示词当成对视频内容的“描述”。这个预设本身就错了。提示词应当是一套电影导演语言。你在定义主体行为、设计镜头语言、控制布光物理逻辑、设定情绪基调、引导运动动态。描述是被动的,导演是主动的,两者产出的结果天壤之别。
这套框架的核心是九层结构:主体、微动作、环境、摄像机行为、灯光设计、视觉风格、情绪基调、物理与运动细节、渲染质量修饰词。每缺少一层,模型的随机性就上升一级。
对比最能说明问题。弱提示词写“a man in a city”,高性能提示词是这样的:
a tired middle-aged man with subtle wrinkles and unshaven beard , wearing a damp dark trench coat , standing under flickering neon signs in a narrow rainy alley , slowly lifting his gaze as raindrops slide down his face , cinematic close-up transitioning into a slow dolly-in shot , shallow depth of field with strong background bokeh , neon reflections shimmering on wet asphalt , soft diffused rim lighting mixed with harsh colored neon highlights , cyberpunk cinematic style , melancholic and introspective mood , slight handheld camera instability for realism , ultra detailed textures , film grain , motion blur , 8k , photorealistic rendering
字数多不是目的,消除歧义才是。长度是结果,不是方法。
在具体执行层面, Matthew 总结了几个被顶级创作者反复验证的策略。
微动作工程( Micro-Motion Engineering )是其中最容易被忽视的一环。 AI 视频最常见的失败不是画面难看,而是“没有东西在有意义地移动”。把“standing still”换成“barely perceptible breathing motion , slight head tilt , subtle eye movement , fabric gently reacting to wind”,同一个场景的生命感会完全不同。
摄像机定义决定了输出是否显得业余。需要明确三件事:运动方式( dolly 推进、 pan 平移、 tracking 跟踪)、构图方式(特写、广角、过肩)、稳定性(手持抖动 vs. 稳定器效果)。没有摄像机逻辑,画面永远像素材,不像电影。
布光是现实感的乘数。平面光照让画面显假,有效的做法是叠加式布光,例如“soft ambient base light combined with strong directional rim lighting and subtle colored reflections”,每一层光都在增加维度感。情绪编码同样影响构图本身。“dark scene”给模型的是颜色指令,“heavy melancholic atmosphere with emotional isolation , visual tension created by empty negative space”给的是构图逻辑。
还有一个被 Matthew 称为“可控混乱”( Controlled Chaos )的进阶技巧:用对比制造层次感,比如将平静主体置于混乱环境、慢动作嵌入高能场景、暖光配情绪上的冷感。这类矛盾感会让画面产生超出预期的深度。
Matthew 整理了五条高性能完整提示词,覆盖不同叙事场景。
情绪钩子场景( Cinematic Emotional Hook )
a young woman with pale skin and expressive eyes standing alone on a deserted street at night , wearing a slightly oversized black coat moving gently in the wind , light rain falling steadily and forming reflections on the asphalt , she slowly raises her head as if realizing something profound , cinematic close-up shot gradually transitioning into a slow forward dolly-in , shallow depth of field isolating her from the blurred neon-lit background , soft diffused key lighting illuminating her face combined with strong colored neon reflections from surrounding signs , subtle rim lighting outlining her silhouette , melancholic and introspective emotional tone , slight handheld camera motion for realism , ultra realistic textures , visible skin detail , film grain overlay , motion blur on rain , 8k cinematic rendering , highly detailed environment
病毒式高能序列( Viral High-Energy Sequence )
a futuristic cyberpunk megacity filled with towering skyscrapers covered in holographic advertisements and glowing neon lights , multiple flying vehicles moving at different speeds between buildings , camera aggressively pushing forward through traffic in a fast-paced tracking shot , dynamic perspective shifts as the camera weaves between structures , strong motion blur emphasizing speed , reflections of neon lights on metallic surfaces and glass panels , dramatic high-contrast lighting with deep shadows and intense highlights , high energy and slightly chaotic atmosphere , cinematic sci-fi style , ultra detailed architecture , volumetric lighting beams cutting through fog , particle effects in the air , 8k ultra realistic rendering , immersive sense of scale and depth
平静叙事场景( Calm but Deep Storytelling )
an elderly fisherman sitting quietly on the edge of a small wooden boat floating on a calm sea at sunrise , gentle waves rocking the boat subtly , he slowly adjusts his fishing net with weathered hands , wide cinematic shot with a slow lateral camera drift , golden hour lighting casting warm soft highlights across the water surface , light mist hovering above the sea creating depth , soft natural color palette , peaceful yet nostalgic emotional tone , photorealistic style , detailed textures on wood and skin , subtle reflections in water , minimal motion creating a meditative rhythm , ultra high detail , film grain , 8k rendering
心理张力场景( Psychological Tension Scene )
a narrow dimly lit corridor inside an abandoned building , peeling paint on the walls and flickering fluorescent lights above , a lone figure standing at the far end partially obscured by shadows , the figure slowly tilts their head unnaturally while remaining otherwise still , camera slowly pushing forward in a suspenseful dolly-in shot , slight handheld instability adding tension , harsh overhead lighting casting long distorted shadows , cold desaturated color grading , tense and unsettling atmosphere , subtle environmental motion like dust particles floating in the air , cinematic horror style , ultra detailed textures , high contrast lighting , film grain , motion blur , 8k photorealistic rendering
自然风光场景( Nature Cinematic Shot )
a vast mountain landscape during early morning with low hanging clouds drifting slowly between peaks , a lone traveler standing on a cliff edge looking over the valley , their cloak gently moving with the wind , wide cinematic aerial shot slowly pulling back to reveal scale , soft golden sunlight breaking through clouds creating volumetric light rays , natural color grading with high dynamic range , epic and awe-inspiring emotional tone , ultra realistic environment detail including rocks , vegetation , and atmospheric depth , smooth camera motion , slight depth haze for realism , 8k cinematic rendering , highly detailed textures
针对高需求视频类型, Matthew 进一步提供了七套适配不同内容方向的完整提示词工程包,涵盖励志电影风( Motivational Cinematic )、冥想内容( Meditation / Inner Peace )、豪华生活方式( Luxury Lifestyle )、暗黑心理悬疑( Dark Psychology / Mystery Hook )、未来科技世界( Futuristic / AI World )、自然与人类情感( Nature + Human Emotion )以及超现实抽象风格( Abstract / Surreal )。
励志病毒短片提示词
a solitary man in his early 30s with a determined yet exhausted expression , slightly sweaty skin and subtle facial tension , standing on the edge of a rooftop overlooking a vast modern city at sunrise , the sky transitioning from deep blue to warm orange tones , wind gently moving his shirt and hair , he slowly clenches his fist and looks forward with renewed determination , cinematic medium-to-close shot transitioning into a slow forward dolly-in , shallow depth of field isolating him from the softly blurred skyline , golden hour lighting casting warm highlights on his face while long shadows stretch behind him , subtle lens flare from the rising sun , inspirational and empowering emotional tone , ultra realistic skin texture and fabric simulation , slight handheld camera movement for authenticity , film grain overlay , motion blur on moving elements , 8k cinematic rendering , high dynamic range , dramatic yet clean color grading
冥想内容提示词
a calm woman sitting cross-legged on a natural rock in the middle of a still lake surrounded by misty mountains , her eyes closed and posture perfectly balanced , long hair gently moving with a soft breeze , the water surface perfectly reflective with subtle ripples expanding outward slowly , camera slowly orbiting around her in a smooth circular motion , wide cinematic shot with gradual push-in creating intimacy , soft diffused morning light with cool blue and pastel tones , volumetric light rays subtly passing through mist , peaceful and deeply introspective emotional atmosphere , extremely smooth motion with no abrupt transitions , ultra realistic environment textures including water reflections and skin detail , minimalistic composition emphasizing silence and space , cinematic depth of field , 8k ultra detailed rendering , soft glow effect , no harsh contrast
豪华生活方式提示词
a luxury modern villa with floor-to-ceiling glass walls overlooking an ocean cliff during sunset , an elegant man in a tailored suit standing on the terrace holding a glass of whiskey , subtle reflections of the sunset visible on the glass surfaces , the ocean waves crashing far below in slow rhythmic motion , camera begins with a wide aerial shot then smoothly transitions into a slow tracking shot approaching the subject from behind , warm golden hour lighting blending with interior ambient lighting from inside the villa , high contrast between warm indoor tones and cool outdoor blues , refined and aspirational emotional tone , ultra realistic materials including glass reflections , marble textures , and fabric detail , cinematic composition with strong leading lines , smooth camera stabilization , slight depth haze over the ocean , filmic color grading , 8k ultra high definition rendering , subtle motion blur
暗黑心理悬疑提示词
a dimly lit room with textured concrete walls and a single chair placed in the center under a flickering overhead light , a mysterious figure in dark clothing sitting motionless with head slightly lowered , face partially hidden in shadow , after a brief still moment the figure slowly lifts their head revealing intense eyes , camera slowly pushing in with a controlled dolly movement , slight handheld instability creating tension , harsh top-down lighting casting deep shadows and emphasizing facial structure , desaturated color palette with cold tones , unsettling and suspenseful emotional atmosphere , subtle environmental motion such as dust particles floating in the air , cinematic horror-thriller style , high contrast lighting , ultra detailed textures on skin and environment , film grain , motion blur on flickering light , 8k photorealistic rendering
未来科技世界提示词
a hyper-advanced futuristic city filled with massive layered architecture , holographic interfaces floating in mid-air , autonomous flying vehicles moving in organized traffic patterns at different altitudes , glowing neon lines outlining buildings and pathways , camera rapidly descending from the sky into the city and then transitioning into a fast-paced forward tracking shot between structures , dynamic perspective shifts enhancing speed and immersion , intense neon lighting with strong reflections on metallic and glass surfaces , volumetric fog interacting with light beams , high energy and immersive emotional tone , cyberpunk sci-fi aesthetic with ultra detailed environment complexity , motion blur emphasizing speed , particle effects in the air , cinematic lighting with deep shadows and bright highlights , 8k ultra realistic rendering , extreme depth and scale
自然与人类情感提示词
a lone traveler standing on a vast open field covered in tall grass moving fluidly with the wind , dramatic cloudy sky above with shifting light breaking through in soft beams , the traveler slowly turns around as if searching for something , wide cinematic shot with a slow backward pull creating a sense of isolation and scale , natural lighting with dynamic contrast between shadowed clouds and illuminated patches of land , emotional tone of solitude and quiet reflection , ultra realistic grass movement simulation , subtle depth haze enhancing distance perception , soft color grading with natural tones , gentle camera motion without abrupt changes , cinematic composition emphasizing negative space , film grain , 8k high detail rendering
超现实抽象风格提示词
a surreal dreamlike environment where gravity behaves inconsistently , large floating geometric structures slowly rotating in mid-air , reflective liquid surfaces flowing upward instead of downward , a human figure walking calmly through the scene unaffected by the distorted physics , camera slowly orbiting and slightly tilting to enhance disorientation , soft glowing ambient lighting combined with sharp highlights on reflective surfaces , color palette shifting gradually between cool and warm tones , ethereal and otherworldly emotional atmosphere , ultra smooth motion with seamless transitions , high detail in reflections and surface textures , cinematic surreal style , depth of field used creatively to isolate elements , 8k ultra detailed rendering , no visible noise , highly polished visual output
Matthew 的核心观点是:更好的提示词不是随机加长,也不是人为增加复杂度,而是结构化、分层次、有意图、具备电影素养。视频表现平庸,不是 Runway 的问题,是摄像机语言、运动设计、布光逻辑、情绪引导这四个要素的集体缺席。