怎么用这个页面
找到和你脑中那个镜头相符的运镜,复制它的提示词片段,然后紧跟在取景和主体之后粘进你的 AI 视频提示词里。每个片段都用了主流文生视频模型读得最稳的措辞,并带上一个速度词,因为「缓慢推进」和「急推变焦」是两条不同的指令,不是同一条的两种写法。每张卡片上的完整示例,展示了这个片段放进一条排序良好的提示词后的样子。
AI 视频生成中的运镜
运镜是 AI 视频提示词里被用错得最多的部分。Veo、Sora、Runway 和 Kling 这类模型,训练素材都是用真实电影词汇描述过的,所以「缓慢左横摇」「跟拍镜头」「摇臂上升」是它们真的会照做的指令,而「史诗级动感镜头」只是噪音,它会自己乱猜着填。
那几个经典的混淆最费重生成次数。横摇是原地转动,整个画面一起扫过,没有视差;而推轨或横移是摄影机真的在移动,前景和背景于是分离开来。变焦改变焦距、把画面压平;推进则穿过空间,透视保持自然。要错了那一个,出来的运动会「不对劲」,却又说不上哪里坏了。
一个镜头一种运镜,是专业默认做法,AI 生成和真实片场都一样。当某个瞬间看起来需要两种运镜时,就生成两段再剪到一起,或者挑出主导的那一种,剩下的交给剪辑。把多条摄影机指令堆进一条提示词,是 AI 素材发糊、发飘最常见的原因。
适合谁
- 用 AI 做电影的人 直接要你脑中那个运镜,而不是把生成次数烧在猜词汇上。
- 电影专业学生 用眼睛学运镜词汇,看它在银幕上真实的样子,而不只是词条解释。
- 视频团队 共享同一份参考,让流水线里的每条提示词都说同一套摄影机语言。
Camera Movement Prompts: the complete guide
It pairs an animated demonstration of every core camera movement with the copy-paste prompt fragment that reliably produces it in Veo, Sora, Runway, Kling, and Seedance.
For this workflow, the central problem is clear: camera movement names are learned from text descriptions, so creators prompt 'dolly' when they mean 'pan' and get footage that moves wrong. Left unresolved, this creates downstream friction and slower decisions. The practical target is you see each movement animated, understand what it does to the frame, and copy the exact prompt phrasing that triggers it in AI video models.
Limitation to keep in mind: Model support for camera language varies: most handle pans, push-ins, and tracking well, while complex moves like orbits and rack focus may need a second generation or a model with explicit camera controls.
Advanced workflow: AI filmmakers pair one movement per shot with the prompt generator, keeping subject descriptions locked across the sequence so the movement changes while the world stays consistent.
Step-by-Step Workflow
- Watch the animations and identify the movement that matches the shot in your head.
- Copy the prompt fragment and paste it into your AI video prompt after subject and setting.
- Expand the example to see the fragment used in a complete, well-ordered prompt.
- Keep one movement per shot: models blur or ignore stacked contradictory camera instructions.
Use Cases By Profile
- AI filmmaker: stop guessing movement vocabulary and prompt the move you actually visualized.
- Film student: learn what each movement does to a frame by watching it, not reading about it.
- Video team: share one reference page so everyone's prompts use the same camera language.
Common Mistakes To Avoid
- Stacking multiple movements in one prompt (pan + orbit + zoom) and getting muddy motion.
- Using 'zoom' when the shot needs a push-in: zoom flattens, dolly moves through space.
- Prompting vague words like 'dynamic camera' instead of a named, specific movement.
Professional Best Practices
- Add a speed word to every movement: 'slow push-in' and 'fast whip pan' behave very differently.
- Pair movement with motivation: 'tracking shot following the subject' beats 'tracking shot' alone.
- If a model ignores the movement, move it earlier in the prompt, right after the shot type.
Treat this tool output as a decision support layer, not a replacement for authorship. Great scripts are remembered for specific choices, emotional precision, and clarity of dramatic movement. Tools help by removing noise so your energy can go where it matters: character, conflict, escalation, and payoff. If you review outcomes after each pass and keep an explicit log of accepted changes, your workflow becomes faster and more predictable from draft to draft. That consistency is exactly what professional collaborators value: fewer surprises, clearer rationale, and a script that evolves with intent.
Extended FAQ
What are the main camera movements in film?
The core vocabulary is pan, tilt, push-in, pull-back, zoom, dolly/truck, tracking shot, orbit or arc, crane or boom, handheld, whip pan, rack focus, and the dutch angle roll. Each one is animated on this page with its matching AI prompt.
Which camera movements work best in AI video models?
Pans, push-ins, pull-backs, tracking shots, and handheld styles are reliably understood by Veo, Sora, Runway, Kling, and Seedance. Orbits, crane moves, and rack focus work but may need a retry or a model with dedicated camera controls.
What is the difference between a pan and a dolly?
A pan rotates the camera in place, sweeping the view with no parallax. A dolly physically moves the camera through space, so foreground and background shift at different speeds. The two animations on this page show the difference instantly.
Why does my AI video ignore my camera movement prompt?
The most common causes are stacking several movements in one prompt, burying the movement at the end of a long prompt, or using vague language. Use one named movement, place it right after the shot framing, and add a speed word.
What is the difference between a zoom and a push-in?
A zoom changes the lens focal length from a fixed position, flattening perspective as it tightens. A push-in physically moves the camera toward the subject, keeping perspective natural. Prompt 'push-in' for cinematic results and 'zoom' only when you want the optical effect.
Can I combine camera movements in one AI prompt?
Generally one movement per shot works best. If the story needs two moves, either pick the dominant one or generate two clips and cut them together, the same way a real production would cover it.