PROMPT RADAR / VIDEO视频叙事

the people who master AI video will be insanely valuable with how good s

the people who master AI video will be insanely valuable with how good some models already are, mastering AI video means being a great prompt engineer, director, architect, designer, and cinematographer... all at once i'

@maxxmalist 2.7k 热度 3.0 万 浏览
编辑先看

它为什么值得进入灵感库

它最有价值的部分,是把一个模糊想法推进到了可以被模型执行的程度。 把一段描述拆成起始状态、动作变化、镜头响应和结束画面,连续性比形容词更重要。

动作与镜头 Prompt

the people who master AI video will be insanely valuable with how good some models already are, mastering AI video means being a great prompt engineer, director, architect, designer, and cinematographer... all at once i'll help you with the first one, here's prompt template for creating highly detailed videos like this:: CAMERA: [Describe the camera type, lens, aspect ratio, framing, movement, and shooting style.] The video should [describe how the camera starts and what happens visually at the beginning]. Maintain [camera/framing behavior] throughout the scene. Camera movement should feel [smooth / handheld / shaky / static / cinematic / improvised / etc.]. Include natural imperfections such as [focus hunting, exposure shifts, imperfect framing, motion blur, reframing, zoom behavior, camera shake, etc.] where appropriate. The camera should [describe whether the camera is visible, mounted, handheld, operated by another person, etc.]. LOOK: [Describe the visual texture, color treatment, lighting response, sharpness, contrast, grain, realism, and overall image quality.] The footage should feel [authentic / cinematic / documentary / polished / raw / nostalgic / futuristic / etc.] and visually consistent with [reference, era, medium, or aesthetic]. STYLE: [Describe the overall creative style and tone of the video.] The scene should feel [natural / spontaneous / dramatic / intimate / energetic / comedic / luxurious / unsettling / etc.]. Characters should behave naturally rather than feeling like they are performing for the camera. Include appropriate [pauses, reactions, gestures, facial expressions, environmental interactions, imperfections, etc.]. CHARACTERS: [Describe each important character.] Character 1: - Age: [age] - Appearance: [appearance] - Hair: [hair] - Clothing: [clothing] - Body type: [body type] - Distinctive features: [features] - Personality / energy: [personality] Character 2: [Repeat structure if needed.] SETTING: [Describe the location, environment, time of day, weather, architecture, objects, background activity, and atmosphere.] The environment should contain [important objects / background elements / environmental details]. Background activity should be [empty / busy / subtle / blurred / chaotic / etc.], while keeping the main subject visually dominant. AUDIO / DIALOGUE: [Describe the sound design, ambient noise, music, dialogue style, and pacing.] Dialogue should sound [natural / conversational / emotional / spontaneous / professional / etc.]. Characters should speak [slowly / quickly / with pauses / casually / confidently / etc.]. Do not make the dialogue sound overly scripted or theatrical unless specifically required. SCENE / ACTION: [Describe the sequence of events in chronological order.] The video begins with: [Opening action / visual hook / first shot.] Then: [Action or beat 1.] [Dialogue or reaction if applicable.] Next: [Action or beat 2.] [Dialogue or reaction if applicable.] Then: [Action or beat 3.] [Dialogue or reaction if applicable.] The scene continues with: [Additional actions, interactions, transitions, or visual beats.] The video ends with: [Final action / visual / dialogue / transition.] --- paste it into an LLM along with your script and ask it to adapt the prompt to your scene
查看原作者内容

不要一次改完所有变量

用 Seedance 复刻时,建议先单独生成稳定首帧,再把动作拆成“开始、变化、收束”三段;每段只保留一个主要运动。

  1. 主体:the people who master AI video will be insanely valuable with how good some models already are, mastering AI video means being a great prompt engineer, director, architect, designer, and cinematographer... all at once i'
  2. 视觉方向:实时 AI 视频案例;查看原帖媒体与作者说明
  3. 镜头与构图:按原帖镜头流程执行;查看原帖媒体与工作流
  4. 交付约束:auto,保留原作者方法与归属;复刻时避免水印、低清、畸变、主体漂移和无关文字。
负向约束

保留原作者方法与归属;复刻时避免水印、低清、畸变、主体漂移和无关文字。

模型建议

Seedance / Veo / Kling。先用原始 Prompt 建立基准结果,再根据模型对自然语言、镜头运动或文字排版的能力做局部改写。

来源与编辑原则

保留作者,也保留判断。

本页收录公开案例并提供中文索引、结构化拆解和复刻建议,不冒充原作者作品,也不替代原帖上下文。

作者:MAX · 收录于 2026年9月5日

访问原始来源

沿着这条线继续看