Seedance 2.0 完整提示词指南
学习如何为 Seedance 2.0 的文生视频、图生视频和参考生视频功能编写有效的提示词。
Seedance 2.0 是一款用于创意视频生成的多模态模型。它可以将文本、图像、视频和音频参考转化为具有明确运动、镜头语言和视听氛围的视频。本指南将说明如何选择合适的生成模式,以及如何通过清晰、可执行的提示词表达你的创作意图。
1. Seedance 2.0 能做什么
1.1 文生视频
文生视频仅根据文本构建场景。你可以描述主体、动作、环境、风格、镜头和声音,再由模型将这一构想发展成完整的动态场景。
这种模式提供最大的创作自由度,适合概念预览、广告创意、氛围短片、奇幻场景,以及尚无视觉素材时的快速探索。由于没有图像或视频作为视觉锚点,提示词的具体程度会直接影响主体设计、空间关系和运动效果。
1.2 图生视频
图生视频可以让静态照片、插画、产品图或概念艺术动起来。输入图像已经确定了主体外观、构图、色彩和基本环境,因此提示词不应只是再次描述图像,而应说明接下来会发生什么,以及场景如何运动。
你可以控制人物动作、微表情、头发和衣物的物理运动,风、雾、光线等环境效果,以及推进、摇摄或固定机位等镜头行为。制作插画、水彩画或 3D 艺术作品的动画时,还可以要求模型保留原始媒介质感,避免结果偏移到另一种视觉风格。
1.3 参考生视频
参考生视频适用于需要更强控制力和一致性的创作。你可以提供图像、视频或音频作为参考,让每项素材分别负责结果中的特定部分:人物外观、产品细节、动作、镜头运动、艺术风格、节奏或情绪发展。
关键在于明确指定每项素材的作用。例如,@image1 可以定义人物外观,@video1 只提供动作与镜头节奏,@audio1 则驱动情绪发展。使用的参考越多,就越需要建立清晰的优先级,避免外观、动作和风格相互冲突。
1.4 选择合适的生成模式
- 你只有一个想法和文字描述: 选择文生视频,从零构建主体、场景和镜头调度。
- 你已经有一张效果出色的首帧、肖像、产品图或艺术作品: 选择图生视频,将提示词重点放在现有图像应如何运动上。
- 你需要复用人物外观、动作、镜头语言、风格或音频节奏: 选择参考生视频,并为每项参考素材指定明确用途。
2. 有效提示词的基础
编写 Seedance 2.0 提示词时,请遵循以下原则:
- 具体、明确: 描述可见或可听的细节,例如主体外观、动作、环境、光线和声音。不要只用“漂亮”或“高级”等模糊词语,而要换成能说明场景实际效果的具体细节。
- 按逻辑组织提示词: 一个可靠的顺序是:主体、动作、环境、风格、镜头和声音。
一个实用的提示词结构如下:
| 要素 | 描述内容 | 示例 |
|---|---|---|
| 主体 / 人物 | 主要主体、外观、服装或关键物件 | 一位身穿深绿色外套、带着一把斑驳木质小提琴的年轻小提琴手 |
| 动作 / 运动 | 主体做什么,包括速度、方向和节奏 | 缓缓穿过车站,然后停在时钟下开始演奏 |
| 环境 / 场景 | 地点、时间、天气、空间纵深和背景活动 | 黎明时分空无一人的老火车站,薄雾从铁轨上飘过 |
| 风格 / 氛围 | 视觉媒介、色彩、光线和情绪基调 | 电影级写实风格,安静而充满希望,冷蓝色阴影与柔和的金色晨光 |
| 镜头 / 构图 | 景别、角度、镜头运动、对焦和画面安排 | 以全景开场,然后缓慢推轨至中近景,浅景深 |
| 音频 / 声音 | 环境声、动作声、音乐或对白 | 远处的火车环境声、轻柔风声、清晰的小提琴独奏,无对白 |
组合成一条完整提示词:
A young violinist in a dark green coat, carrying a weathered wooden violin, walks slowly through an empty old railway station at dawn, then stops and begins to play beneath the clock. Light fog drifts across the tracks. Cinematic realism, quiet and hopeful, with cool blue shadows and soft golden morning light. Begin with a wide shot, then make a slow dolly-in to a medium close-up with shallow depth of field. Distant train ambience, soft wind, clear solo violin, no dialogue.- 使用精准关键词: 用
soft lighting、watercolor style或slow tracking shot等短语定义质感、艺术风格或镜头语言。避免堆砌重复或相互矛盾的词语。 - 写明重要排除项: 对关键限制使用负向指令,例如“no cuts”“do not change the character's appearance”或“do not make the image photorealistic”。只保留真正重要的限制。
- 明确分配参考素材的职责: 上传图像、视频或音频文件后,说明它负责控制外观、动作、镜头行为还是音乐。
- 测试并迭代: 第一次结果未必完美。每次只调整少量词语、细节或优先级,以便判断哪些指令改善了输出。
3. Seedance 2.0 提示词框架
3.1 文生视频生成
文生视频是最直接的 Seedance 2.0 工作流。由于没有图像或视频作为视觉锚点,提示词的结构和具体程度尤其重要。
提示词
A solitary meteorologist in a bright orange weather suit stands on a black volcanic ridge as a vast thunderstorm approaches across the ocean. She raises a handheld sensor into the wind; her coat and loose straps whip violently while sheets of rain sweep across the rocks. A distant lightning strike illuminates the cloud layers, and she turns toward the flash with a focused expression. Photorealistic cinematic drama, cold steel-blue palette, wet reflective textures, strong backlight through the rain. Start with a wide establishing shot, then track slowly around her to a low-angle medium shot as the lightning flashes. Deep wind, rolling thunder, rain striking fabric and stone, no dialogue.输出 · 视频
- 以主体开头: 将最重要的人物、产品或物件放在提示词开头,让模型迅速建立视觉焦点。
- 清楚描述运动: 运动是视频的核心。用
slowly panning left、rapidly spinning或gently swaying等短语明确运动的类型、速度和方向。 - 定义节奏: 如果时序很重要,可将片段描述为
a slow, meditative sequence或a fast-paced, high-energy montage。 - 定义光线: 说明光源、方向和质感,例如
soft, diffused morning light、harsh neon backlighting或flickering candlelight casting warm shadows。 - 加入氛围:
tension-filled、whimsical、melancholic或euphoric等词有助于建立情绪基调。
3.2 图生视频生成
提供静态照片、艺术作品或概念图后,提示词会告诉 Seedance 2.0 如何让现有画面鲜活起来。
与文生视频不同,你无需用文字重新搭建场景。重点描述接下来会发生什么,以及图像的每一层应如何运动。
分层设计运动
提示词
Animate with fine red dust sweeping across the cracked ground in the foreground. The astronaut's loose fabric straps flutter gently in a steady wind, and faint condensation gathers along the inside edge of the visor. In the background, red warning lights flicker on the abandoned outpost, and two reconnaissance drones circle lazily above the antenna towers. Apply a slow, contemplative camera push-in toward the astronaut. The mood is lonely and monumental, with a muted rust-and-cyan color grade and soft, diffused sunset lighting.输入 · @image1

输出 · 视频
有效的图生视频提示词会将场景视为多个可以独立运动的层次。
初学者常见的错误是只描述主体动作,而忽略环境变化。让前景、主体和背景协同运动,通常能产生更自然、更具电影感的结果。
- 前景: 最靠近镜头的元素,例如飘动的树叶、闪烁的烛光或水面涟漪。
- 中景: 主要主体及其核心动作,例如人物缓缓转头或马匹转移重心。
- 背景: 营造纵深的元素,例如移动的云层、远处的旗帜或主体后方远处行走的人群。
三个层次不需要具有同等强度。先确保主体动作清晰,再以更细微的前景和背景运动加以衬托。
用微动作传达情绪
当图像中有人物或面部特写时,细微的表情与姿态变化可以让主体显得鲜活,同时避免引入不稳定的运动。
细小的眼神移动、轻微眯眼、可见的呼吸或随风摆动的衣领,往往比夸张表演更能传达情绪。
提示词
Animate with a barely perceptible shift in the fisherman's gaze—his eyes slowly tracking something distant on the horizon. A faint squint tightens around his eyes. His jacket collar flutters softly. Waves reflect subtly in his eyes. The camera remains completely static, locked off. The atmosphere is deeply contemplative and nostalgic, desaturated with warm tones with soft coastal light.输入 · @image1

输出 · 视频
保留抽象与艺术风格
图生视频不仅能让照片动起来,也能为绘画、插画和概念艺术制作动画。
对于这类参考,请明确要求保留原始媒介的笔触、边缘、色彩、纹理和材质。这样可以防止动画变得过度写实或在视觉上不一致。
提示词
Animate this scene while fully preserving the dreamy watercolor storybook aesthetic—soft, luminous color washes, delicate painterly textures, glowing gold details, and diffused starlight throughout. The little boy gently plays the harp, his fingers softly plucking the strings as they shimmer and vibrate with golden light. White birds circle gracefully around the harp, while swallows glide and flutter across the starry sky in smooth, flowing paths. The animation should feel hand-crafted, poetic, and delicate, never sharp or digital. A gentle, whimsical atmosphere with celestial blue, soft white, and warm golden tones.输入 · @image1

输出 · 视频
图生视频关键词参考
这些不是固定公式。请根据所需的运动和氛围进行组合。
| 类别 | 实用关键词 |
|---|---|
| 运动质感 | gently、barely perceptible、slowly drifting、rhythmically swaying、subtly rippling |
| 氛围 | mist rolling in、particles of dust、heat haze、soft bokeh、volumetric light rays |
| 人物生命感 | micro-expression shift、eyes slowly tracking、breath visible、hair softly lifted by wind |
| 镜头 | locked off、slow push-in、subtle drift、gentle handheld sway、rack focus |
| 风格保留 | maintain painterly texture、preserve film grain、honor the original color palette |
3.3 参考生视频生成
保留与转换法
参考生视频提示词可以明确分成两部分:保留与转换。这样能直接告诉 Seedance 2.0 哪些内容必须保持一致、哪些内容应当改变,而不是让模型自行猜测。
提示词
(Preserve) Retain all original movement, choreography, timing, and body posture of the dancer exactly as they appear in the source video. Maintain the original camera angle and framing throughout.
(Transform) Re-stylize the entire visual environment as an ethereal, otherworldly forest glade. Replace the studio floor with a carpet of luminous, floating flower petals. Surround the dancer with slow-moving fireflies and drifting luminescent spores. The dancer's costume should transform into a flowing, translucent gown that trails light. Apply a dreamlike, fantasy aesthetic with soft teal and lavender tones, volumetric god rays filtering through ancient trees. Film grain texture, cinematic quality.输入 · @video1
输出 · 视频
风格迁移:定义新的视觉语言
当你希望保留源视频内容、但改变整体风格时,不要只给出一个宽泛的风格名称,而要用具体词语描述目标美学。明确色彩、光线、材质、画面质感和年代细节。
同时说明必须保留哪些源视频元素,例如主体动作、表演时序、对白时序或镜头运动。
提示词
Preserve all dialogue timing, gestures, and the camera position. Re-stylize the entire scene in the visual language of a Studio Ghibli animated feature—soft, hand-drawn cel animation aesthetic, warm and richly textured backgrounds, characters rendered with expressive Ghibli-style proportions. The café transforms into a charming, vintage European bakery with afternoon sunlight streaming through lace curtains. Palette is warm, creamy, and inviting. Gentle ambient sounds of clinking cups and soft piano music implied in the visual atmosphere.输入 · @video1
输出 · 视频
多模态融合
多模态生成可以在一次请求中组合图像、视频和音频参考。这种自由也增加了复杂度:每项素材都可能带来不同的风格、节奏、色彩或情绪基调。目标是建立统一的创作方向,避免各项参考彼此竞争。
建立创作优先级
可以把输入素材想象成电影制作中的不同部门。一项参考定义视觉身份,另一项编排动作和镜头行为,还有一项控制情绪节奏。每项素材都应只有一个清晰职责。
提示词
@image1 is the primary visual authority—the protagonist’s face, hairstyle, blue hair streak, cybernetic eye implant, black tactical clothing, illuminated cyan details, boots, and messenger bag must remain exactly consistent throughout the entire video. Do not redesign, replace, or simplify any part of her appearance.
@video1 serves exclusively as the movement, parkour choreography, body-mechanics, and camera reference. Apply its exact sprinting rhythm, barrier vault, wall run, landing impact, low slide, and final acceleration to the protagonist from @image1. Preserve the original sequence, timing, spatial direction, tracking shots, camera orbit, and low floor-level camera movement, but do not carry over the gray training outfit, stunt performer’s identity, warehouse, or any other visual element from @video1.
@audio1 sets the emotional rhythm and editing intensity of the entire sequence. During the restrained opening pulses, begin with controlled running and a smooth low-angle tracking shot. As the percussion builds, increase the protagonist’s speed, environmental motion, and camera energy. Synchronize the vault, wall push, slide, and strongest camera movements with the major rhythmic accents. At the musical drop, reveal the chase at full intensity and finish on the final impact.输入 · @image1

输入 · @video1
输入 · @audio1
输出 · 视频
融合法:等权重参考
当两项或更多素材确实需要融合成一个新的视觉世界时,请说明它们如何结合,而不是指定其中一项为主导参考。
提示词
Fuse the visual identities of @image1 and @image2 equally into a single, cohesive world—a retro-futurist city that exists at the intersection of 1930s art deco grandeur and contemporary neon Tokyo nightlife. Neither should dominate; the architecture carries the geometric elegance of @image2 while glowing with the saturated neon palette and wet-reflective streets of @image1. Animate a slow, gliding aerial camera drift through this world, unhurried and contemplative. Let @audio1 dictate the pace entirely—every camera movement should feel as languid and swinging as the jazz rhythm. The atmosphere is nostalgic, mysterious, and quietly beautiful.输入 · @audio1
输入 · @image1

输入 · @image2

输出 · 视频
以音频作为主要驱动力
音乐和声音设计可以从头到尾控制视频结构。请说明场景应如何随声音发展作出反应:安静段落保持克制,音乐渐强时增加环境运动,并在高潮处呈现最强烈的视觉变化。
提示词
Let @audio1 be the architect of this entire video. Begin in near-silence: a static, locked-off shot of the lighthouse from @image1—still, barely animated, only the faintest movement of stormy clouds. As the orchestral score begins to swell, incrementally increase the intensity of the environment—waves grow larger, lightning begins to flash in the distance, the wind picks up, the lighthouse beam begins to rotate. By the time the score reaches its full crescendo, the scene should be a breathtaking storm in full fury—crashing waves, torrential rain, dramatic lightning strikes illuminating the cliff face, the lighthouse beam cutting through the chaos. The visuals and music must feel inseparable, as if one created the other. Cinematic, photorealistic, deeply dramatic.输入 · @audio1
输入 · @image1

输出 · 视频
4. 总结
优秀的提示词不是一堆形容词,而是对创作职责的清晰划分:主体是谁或是什么、会发生什么、环境如何响应、镜头如何观察场景,以及声音如何塑造情绪发展。
文生视频从零创造画面。图生视频让现有画面鲜活起来。参考生视频则通过明确的素材职责来锁定外观、动作、风格和节奏。
先从一个清晰的视觉想法开始,再以小步迭代的方式调整动作、镜头行为和氛围。相比试图一次性规定所有细节,明确的意图更加可靠。