我测试了大家在 AI 视频提示（Sora、Veo、Runway、Kling）中都在用的 10 个"万能短语"。

I tested 10 "power phrases" everyone uses in AI video prompts (Sora, Veo, Runway, Kling). Only half of them actually work. Here's the breakdown, ranked by real evidence: ✅ ACTUALLY WORKS → "Shallow depth of field" - consistently produces bokeh across tools → "Volumetric light / god rays" - strong, repeatable atmospheric effect → "Practical lighting, motivated source" - reliably improves realism → "Slow push in, 4-second clip" - matches how these models handle short, restrained motion best → "Diegetic sound only" - genuinely improves audio realism in tools with native sound (Veo, Sora) ⚠️ SOMEWHAT WORKS (style nudge, not guarantee) → "Shot on ARRI Alexa" - biases toward a cinematic look, doesn't replicate real camera data → "Anamorphic lens, flares" - often adds flares, but inconsistently → "Skin pore detail, subsurface scattering" - helps at high resolution, weak at lower settings ❌ WEAK OR PLACEBO → "24fps, natural motion blur" - most models render at a fixed frame rate regardless of what you type → "f/1.8" as a numeric value - no verified correlation with output; the word "shallow" does the real work → "In the style of [cinematographer]" - inconsistent, model dependent, sometimes ignored entirely THE ACTIONABLE TAKEAWAY: - Stop stacking jargon. - Use plain, physical language: - Describe the shot's intent, not camera specs - Name the lighting behavior, not the gear - Keep motion instructions simple and short - Skip numeric camera settings entirely Test one modifier at a time so you know what's actually working What's a prompt phrase you swear by? Curious if it holds up.

我测试了大家在 AI 视频提示（Sora、Veo、Runway、Kling）中都在用的 10 个"万能短语"。 其中只有一半真正有效。 以下是按真实证据排名的分类： ✅ 真正有效 → "Shallow depth of field"（浅景深）—— 在各工具中都能稳定生成虚化效果 → "Volumetric light / god rays"（体积光 / 神圣光束）—— 强烈且可复现的氛围效果 → "Practical lighting, motivated source"（实用光、动机光源）—— 可靠地提升真实感 → "Slow push in, 4-second clip"（缓慢推进，4 秒片段）—— 最贴合这些模型处理短而克制运镜的方式 → "Diegetic sound only"（仅使用画内声音）—— 在具备原生音效的工具（Veo、Sora）中能真正提升音频真实感 ⚠️ 部分有效（属于风格倾向，而非保证） → "Shot on ARRI Alexa"（使用 ARRI Alexa 拍摄）—— 偏向电影感外观，但无法复现真实摄影机数据 → "Anamorphic lens, flares"（变形镜头、光斑）—— 常常会增加光斑，但效果不稳定 → "Skin pore detail, subsurface scattering"（毛孔细节、次表面散射）—— 在高分辨率下有帮助，在较低设置下效果较弱 ❌ 效果弱或安慰剂 → "24fps, natural motion blur"（24fps，自然动态模糊）—— 大多数模型无论你输入什么，都会以固定帧率渲染 → 数值形式的 "f/1.8"—— 与输出之间没有经过验证的相关性；真正起作用的是"shallow"这个词 → "In the style of [cinematographer]"（风格模仿 [某位摄影指导]）—— 效果不一致，依赖模型，有时会被完全忽略 可执行的要点： - 停止堆砌术语 - 使用朴素、具象的语言 - 描述镜头意图，而非摄影机参数 - 说出光线的行为，而不是器材名称 - 保持运镜指令简短 - 完全跳过数值化的摄影机设置 一次只测试一个修饰词，这样你才能知道哪个真正有效 你最信赖的提示短语是什么？很好奇它是否经得起检验。