文章
如何写出更好的 Seedance 2.0 视频 Prompt
Seedance 2.0 是目前最强大的多模态视频模型之一:你可以在单次生成中组合文本、图片、参考视频和音频。但关键在于,效果很大程度上取决于你怎么写 Prompt — 尤其是如何告诉模型每个上传文件的用途。
本指南梳理在 HiArt 上能稳定出片的语法与写法,无论你是做 6 秒产品循环,还是 15 秒叙事短片。
可以上传什么
写 Prompt 前先了解限制 — 提前规划素材能少返工。
输入限制
输入类型 | 数量限制 | 格式 | 最大体积 |
|---|---|---|---|
图片 | ≤ 9 | jpeg, png, webp, bmp, tiff, gif | 每张 30 MB |
视频 | ≤ 3 | mp4, mov | 每个 50 MB,总时长 2–15s |
音频 | ≤ 3 | mp3, wav | 每个 15 MB,总时长 ≤ 15s |
文本 | 自然语言 Prompt | — | — |
合计 | ≤ 12 个文件 | — | — |
输出规格
- 时长:4–15 秒(自行选择)
- 自动生成音效与背景音乐
- 分辨率:480p–720p
补充说明
- 上传素材中的真实人脸会因合规要求被拦截 — 请改用风格化或非写实参考。
- 使用参考视频的生成成本略高于纯图片生成。
- 优先上传对画面与节奏影响最大的文件。
@ 引用系统
每个上传素材都会对应一个槽位 —
@Image1、@Video2、@Audio1 等。模型不会自动推断每个文件的用途,你必须在 Prompt 里明确说明。@Image1 @Image2 @Image3 @Video1 @Video2 @Audio1
常见引用角色
用途 | 示例 |
|---|---|
首帧 | @Image1 as the first frame |
尾帧 | @Image2 as the last frame |
角色外观 | @Image1's character as the subject |
场景 / 背景 | scene references @Image3 |
镜头运动 | reference @Video1's camera movement |
动作 / 编排 | reference @Video1's action choreography |
特效 / 转场 | completely reference @Video1's effects and transitions |
节奏 / 节拍 | video rhythm references @Video1 |
背景音乐 | BGM references @Audio1 |
音效 | sound effects reference @Video3's audio |
可以在一个 Prompt 中叠加多个角色:
@Image1's character as the subject, reference @Video1's camera movement and action choreography, BGM references @Audio1, scene references @Image2
如何组织 Prompt
效果好的 Prompt 通常按以下顺序展开:
[主体] + [场景] + [动作/运动] + [镜头] + [时间分段] + [特效/转场] + [音频] + [风格/氛围]
超过 10 秒的片段,建议把动作拆成带时间节点的节拍 — 模型会可靠得多:
0–3s: 开场场景、镜头、动作 3–6s: 中段发展 6–10s: 高潮或关键动作 10–15s: 收尾镜头、最终文字或品牌标识
镜头语言速查
Seedance 对标准影视术语响应很好。把运动方式与景别组合使用,控制会更精细。
运动方式
术语 | 效果 |
|---|---|
Push in / Slow push | 镜头向主体推进 |
Pull back | 镜头远离主体 |
Pan left/right | 水平摇镜 |
Tilt up/down | 垂直摇镜 |
Track / Follow shot | 镜头跟随主体 |
Orbit / Revolve | 镜头环绕主体 |
One-take / Oner | 一镜到底,无剪辑 |
景别
术语 | 画面范围 |
|---|---|
Extreme close-up | 眼睛、嘴部或局部细节 |
Close-up | 面部占满画面 |
Medium shot | 腰部以上 |
Full shot | 全身 |
Wide / Establishing | 完整环境 |
实用示例
保持角色一致
用参考图锚定主体,再按节拍描述场景:
The man in @Image1 walks tiredly down the hallway, slowing his steps, finally stopping at his front door. Close-up on his face — he takes a deep breath, adjusts his emotions, replaces the weariness with a relaxed expression. After entering, his daughter and a pet dog run to greet him. The interior is warm and cozy. Natural dialogue throughout.
复刻参考视频的镜头
Reference @Image1's male character. He is in @Image2's elevator. Completely reference @Video1's camera movements and facial expressions. Hitchcock zoom during the fear moment, then orbit shots of the elevator interior. Elevator doors open, follow shot walking out. Exterior scene references @Image3.
延长已有片段
生成时长需与延长长度一致(延长 5 秒 → 选择 5 秒输出):
Extend @Video1 by 15 seconds. 1–5s: Light and shadow slide across a wooden table through venetian blinds. 6–10s: A coffee bean drifts down from the top of frame. Camera pushes in until the screen goes black. 11–15s: Text appears — "Lucky Coffee", "Breakfast", "AM 7:00-10:00".
产品展示
Deconstruct the reference image. Static camera. Hamburger suspended and rotating mid-air. Ingredients gently separate while maintaining shape. Bun top, lettuce, tomato, beef patties, melting cheese, bun base — all descend and reassemble into a complete burger. Cheese melts and drips, lettuce and tomato dewdrops glisten.
卡点剪辑
@Image1 @Image2 @Image3 @Image4 @Image5 @Image6 @Image7 — match the keyframe positions and overall rhythm of @Video1 for beat-synced cuts. More dynamic character movement. Dreamlike visual style with strong tension. Adjust shot sizes and lighting to follow the music.
风格与音频修饰语
在 Prompt 末尾追加短句,引导画面与声音:
- 画面:
Cinematic quality, film grain, shallow depth of field·2.35:1 widescreen, 24fps·Anime style/Photorealistic - 氛围:
Tense and suspenseful·Warm and healing·Epic and grand - 音频:
Background music: grand and majestic·Beat-synced transitions matching music rhythm·Voice tone reference @Video1
快速工作流
- 确定视频类型 — 广告、短剧、MV、产品、教程等。
- 收集图片、参考视频和音频,并记录每个素材对应的槽位。
- 为每个
@引用分配角色 — 首帧、镜头、BGM 等。 - 按顺序写场景:主体 → 环境 → 动作 → 镜头 → 时间分段。
- 补充风格与音频修饰语,核对文件数量与时长限制。
- 生成、审片,并在下一轮迭代中收紧表述模糊的句子。
常见误区
- 引用含糊 — 只写「reference @Video1」,却不说明是镜头、动作还是节奏。
- 镜头指令冲突 — 同一段里既写固定机位又写环绕镜头。
- 短片段塞太多 — 4 秒里切三个场景,往往很难出好效果。
- 上传了却没写 — 传了五张图,Prompt 里只提到两张。
- 没有音频方向 — 声音设计是最快的提质手段之一。
- 时长不匹配 — 写了 15 个节拍,输出却只选了 5 秒。
可直接复制的模板
15 秒产品广告
Reference @Video1's editing style and camera transitions. Replace @Video1's product with @Image1 as the hero product. 0–3s: Product enters with dynamic rotation, close-up on texture and logo. 4–8s: Angle transitions — front, side, back — with highlight scanning light. 9–12s: Product in lifestyle usage context. 13–15s: Hero shot, brand tagline appears, music builds to resolution. Sound: Reference @Video1's background music. Add product interaction SFX.
15 秒短剧
Scene (0–5s): Close-up on reddened eyes, finger pointing accusingly, tears streaming. Dialogue (A, choking): "What exactly are you trying to take from me?" Scene (6–10s): Other character trembles, holds up evidence, steps forward. Dialogue (B): "I'm not deceiving you! This is what he entrusted to me!" Scene (11–15s): Evidence revealed. A freezes — anger shifts to shock. Sound: Urgent piano, static interference, sobbing, button click.
13 秒舞蹈片段
Have the character in @Image1 replicate the dance moves and beat-synced music from @Video1. Generate a 13-second video. Movements smooth, no stuttering or freezing.
准备好试了吗?在 HiArt canvas 打开视频节点,选择 Seedance 2.0,上传参考素材,并粘贴本指南中的 Prompt。先从简单开始 — 一张图、一个清晰动作、一种镜头运动 — 再根据模型表现逐步叠加复杂度。