Cinematic Directing Techniques — Multi-Shot Prompt
演示视频
原帖未提供效果图
提示词
The difference between “AI slop” and “this looks directed” is literally 2–3 prompt changes. Seedance 2 Global access api :- If you are just looking for an app to run Seedance 2 without business email and geo restrictions check out VadooAI The “4-Part Prompt Formula” 🎬 Most people write vibes. Seedance wants structure. Use this every time: Subject → Action → Camera → Style If you skip camera → you get random motion If you skip action → you get stiff clips 2. The “Camera is King” rule 🎥 Seedance is basically a camera simulator. Instead of: “girl walking in city” Say: “tracking shot, slow dolly-in, 35mm lens, shallow depth of field” That alone upgrades output quality massively 3. The “Don’t Describe the Image” hack 🖼️ If you’re using a reference image: DO NOT re-describe it Only describe motion + changes Otherwise the model “reinterprets” your image and ruins it 4. The “Motion Intensity” trick ⚡ Seedance doesn’t infer speed. “car drives” = boring “car accelerates aggressively, motion blur, tires screeching” = cinematic You have to explicitly define energy levels
中文译文
“AI 垃圾感”与“这看起来像被导演过”之间的差别,其实就只是 2–3 处提示词改动。 Seedance 2 全球访问 api:- 如果你只是想找一个无需企业邮箱、不受地域限制就能运行 Seedance 2 的应用,可以看看 VadooAI “四段式提示词公式” 🎬 大多数人写的是感觉。Seedance 要的是结构。 每次都套用这个: 主体 → 动作 → 镜头 → 风格 跳过镜头 → 你会得到随机的运动 跳过动作 → 你会得到僵硬的片段 2. “镜头为王”法则 🎥 Seedance 本质上就是一台摄影机模拟器。 不要写:“女孩走在城市里” 而要写:“跟拍镜头,缓慢推镜,35mm 镜头,浅景深” 仅这一项就能大幅提升输出质量 3. “不要描述图像”技巧 🖼️ 如果你在用参考图: 不要再去描述它 只描述运动 + 变化 否则模型会“重新诠释”你的图像并把它毁掉 4. “运动强度”诀窍 ⚡ Seedance 不会自行推断速度。 “车在行驶” = 无聊 “汽车猛烈加速,运动模糊,轮胎尖叫” = 电影感 你必须明确地定义能量强度
溯源
- 原作者
- noman23761
- 发布日期
- 2026-04-07(由推文 ID 反解)
- 原始出处
- 查看原帖 ↗
- 收录来源
- EvoLinkAI/awesome-seedance-2.5-prompts · 2041409914954973216 · Apache-2.0