KIVIDE AI使用高级模型生成 AI 视频与图像，无需信用卡。
用一句自然语言描述你的创意，选择目标模型，即可得到结构化的提示词——分为镜头、主体、动作、灯光、运镜、风格与画质七段。免费，无需注册。
生成视频提示词
目标模型
Write it like a shot brief with an audio cue; camera and sound are both first-class.
结构化提示词
VeoModel dialects
每个模型训练时见过的描述方式不同，因此「听得懂」的词也不同。Veo 把镜头与声音当作一等要素，点明镜头焦段和一句音频提示，效果远胜堆砌形容词。Kling 偏好单一、清晰的物理动作，并会响应否定词；但把多个动作堆在一起，画面就会变得像橡胶般扭曲。Seedance 期待有顺序的分镜，如果时间线含糊，它会塌缩成单个镜头。Sora 读的是叙事性的散文，基本忽略否定词，所以你要描述「应该发生什么」，而不是「不要发生什么」。Runway 以运镜为先，最怕把运镜淹没在冗长的文学化长句里。Wan 则很直白：具体名词与动作有效，隐喻无效。把同一段泛泛的文案复制到所有模型上，是好创意渲染效果差的头号原因。
六个主流模型分别希望被怎样描述。请把它当作使用场景指引，而非质量排名——这些模型更新频繁。
| 模型 | 最擅长 | 建议长度 | 否定词 | 常见坑 |
|---|---|---|---|---|
| Veo | Cinematic realism, native audio | Medium–long (30–60 words), with an explicit audio line | 部分支持 | Describe sound explicitly and add “(no subtitles)” or the model may render caption text. |
| Kling | Character motion, physical realism | Medium (20–40 words), one clear action | 支持 | Too many simultaneous actions make the motion rubbery — keep to one primary movement. |
| Seedance | Multi-shot sequencing, timelines | Long (40–80 words), ordered beats | 支持 | Vague sequencing makes the timeline collapse into one shot — order the beats explicitly. |
| Sora | Complex scenes, narrative | Long (50–100 words), prose | 不可靠 | List-like fragments read as flat — write full sentences with a sense of story. |
| Runway | Image-to-video, camera control | Short (10–25 words), camera-first | 部分支持 | Long literary prompts dilute the camera instruction; keep it tight and motion-led. |
| Wan | Cost-efficient clips, product demos | Medium (20–40 words), concrete clauses | 部分支持 | Abstract or poetic language is ignored — name concrete objects and motions. |
覆盖最常见的创意类型，每个都标注了适配模型。下方每条提示词都以纯文本存在于本页 HTML 中，即使禁用 JavaScript 也能复制。
黄昏时分，一名孤独的宇航员走过红色沙漠
Shot: Single shot, 16:9, ~5 seconds Subject: 黄昏时分，一名孤独的宇航员走过红色沙漠 Action: 黄昏时分，一名孤独的宇航员走过红色沙漠 Lighting: Soft directional key light, balanced fill Camera: Slow dolly-in, 35mm lens, shallow depth of field Style: Photoreal, cinematic grade, natural motion Quality: Sharp focus, natural film grain, no text or watermarks
一只玻璃香水瓶在黑色台面上旋转
Shot: Single shot, 16:9, ~5 seconds Subject: 一只玻璃香水瓶在黑色台面上旋转 Action: 一只玻璃香水瓶在黑色台面上旋转 Lighting: Soft directional key light, balanced fill Camera: Camera pushes in from a low angle Style: Clean commercial look, controlled motion Quality: Sharp focus, natural film grain, no text or watermarks
创始人在明亮的办公室里讲解产品
Shot: Single shot, 16:9, ~5 seconds Subject: 创始人在明亮的办公室里讲解产品 Action: 创始人在明亮的办公室里讲解产品 Lighting: Soft directional key light, balanced fill Camera: Medium tracking shot, eye-level, natural handheld weight Style: Realistic, physically believable motion Quality: Sharp focus, natural film grain, no text or watermarks
少年在雨中拔出闪光的剑
Shot: Two-shot sequence, 16:9, ~8 seconds total Subject: 少年在雨中拔出闪光的剑 Action: 少年在雨中拔出闪光的剑 Lighting: Soft directional key light, balanced fill Camera: Shot 1: wide establishing; Shot 2: medium close-up Style: Sequenced, continuity across shots Quality: Sharp focus, natural film grain, no text or watermarks
夜晚的霓虹街道与复古跑车
Shot: Single shot, 16:9, ~5 seconds Subject: 夜晚的霓虹街道与复古跑车 Action: 夜晚的霓虹街道与复古跑车 Lighting: Soft directional key light, balanced fill Camera: Static medium shot with a gentle pan right Style: Clean, efficient, product-demo clarity Quality: Sharp focus, natural film grain, no text or watermarks
电动车沿滨海公路飞驰
Shot: Single shot, 16:9, ~5 seconds Subject: 电动车沿滨海公路飞驰 Action: 电动车沿滨海公路飞驰 Lighting: Soft directional key light, balanced fill Camera: Slow dolly-in, 35mm lens, shallow depth of field Style: Photoreal, cinematic grade, natural motion Quality: Sharp focus, natural film grain, no text or watermarks
日出时雾气森林中的瀑布
Shot: Single continuous moment, 16:9, ~10 seconds Subject: 日出时雾气森林中的瀑布 Action: 日出时雾气森林中的瀑布 Lighting: Soft directional key light, balanced fill Camera: Smooth crane rise into a wide two-shot Style: Narrative, filmic, emotionally grounded Quality: Sharp focus, natural film grain, no text or watermarks
一艘飞船从星云中驶出
Shot: Single continuous moment, 16:9, ~10 seconds Subject: 一艘飞船从星云中驶出 Action: 一艘飞船从星云中驶出 Lighting: Soft directional key light, balanced fill Camera: Smooth crane rise into a wide two-shot Style: Narrative, filmic, emotionally grounded Quality: Sharp focus, natural film grain, no text or watermarks
一个人在桌前拆箱耳机，手机拍摄风格
Shot: Single shot, 16:9, ~5 seconds Subject: 一个人在桌前拆箱耳机，手机拍摄风格 Action: 一个人在桌前拆箱耳机，手机拍摄风格 Lighting: Soft directional key light, balanced fill Camera: Camera pushes in from a low angle Style: Clean commercial look, controlled motion Quality: Sharp focus, natural film grain, no text or watermarks
温馨厨房里一碗拉面冒着热气
Shot: Single shot, 16:9, ~5 seconds Subject: 温馨厨房里一碗拉面冒着热气 Action: 温馨厨房里一碗拉面冒着热气 Lighting: Soft directional key light, balanced fill Camera: Medium tracking shot, eye-level, natural handheld weight Style: Realistic, physically believable motion Quality: Sharp focus, natural film grain, no text or watermarks
一只金毛犬在海滩上奔跑
Shot: Two-shot sequence, 16:9, ~8 seconds total Subject: 一只金毛犬在海滩上奔跑 Action: 一只金毛犬在海滩上奔跑 Lighting: Soft directional key light, balanced fill Camera: Shot 1: wide establishing; Shot 2: medium close-up Style: Sequenced, continuity across shots Quality: Sharp focus, natural film grain, no text or watermarks
镜头滑过一座现代玻璃住宅
Shot: Single shot, 16:9, ~5 seconds Subject: 镜头滑过一座现代玻璃住宅 Action: 镜头滑过一座现代玻璃住宅 Lighting: Soft directional key light, balanced fill Camera: Static medium shot with a gentle pan right Style: Clean, efficient, product-demo clarity Quality: Sharp focus, natural film grain, no text or watermarks
点明画面里的谁/什么，并加一个辨识细节——「穿着磨损白色宇航服的孤独宇航员」胜过「一个孤独的人」。主体含糊，模型就会自行编造你之后得费力修正的细节。一条提示词只保留一个主体，焦点才清晰。
直接写运镜方式和焦段：推轨、升降、手持跟拍、35mm、浅景深。像 Runway 这类以运镜为先的模型，需要它出现在第一个分句；放在最后就被淹没。单个镜头里不要叠加两种运镜。
描述光源本身而非情绪：「温暖的黄金时刻主光，长长的阴影」或「侧面低照度霓虹」。灯光对真实感的影响超过任何其他字段，并与创意里的时间设定直接呼应。
Kling 与 Seedance 对明确的回避清单反应良好（多出的肢体、扭曲的文字、重复物体）。Sora 基本忽略否定词，对它应改为描述正确结果。拿不准时，把「不要模糊」改写为「清晰、对焦准确」。
免费。你可以直接构建、复制、重新生成提示词，无需注册。真正生成视频需要账号，新账号注册即送免费积分。
生成器覆盖六个主流模型家族——Veo、Kling、Seedance、Sora、Runway 与 Wan——每个都有自己的提示词偏好、长度建议与否定词行为。Kivide 工作室运行其背后的生产级引擎，你可以在那里切换模型而无需重写提示词。
是的，这正是本页的价值所在。同一个创意写给 Runway（简短、运镜优先）会在 Seedance（分镜顺序）上表现不佳，反之亦然。选择上方的目标模型，结构会自动调整。
只在模型真正响应时才用。Kling 与 Seedance 会对明确的回避清单作出反应；Sora 基本忽略否定词，应改为描述期望的结果。上方速查表标明了各自的支持情况。
以运镜控制为主的模型（如 Runway）要短，10–25 个词；分镜模型（如 Seedance）40–80 个词；叙事模型（如 Sora）50–100 个词。各模型建议长度已列于表中。
