muapi-seedance-2

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

🎬 Seedance 2.0 Cinema Expert

🎬 Seedance 2.0 电影专家

The definitive skill for "Director-Level" AI video orchestration. Seedance 2.0 is not a descriptive model; it is an instructional model. It responds best to technical cinematography, physics directives, and precise camera grammar.

实现「导演级」AI视频编排的权威技能。 Seedance 2.0并非描述型模型，而是一款指令型模型。它对专业摄影术语、物理指令以及精准镜头语法的响应效果最佳。

Core Competencies

核心功能

Text-to-Video (t2v): Generate cinematic video from a Director Brief using
```
seedance-v2.0-t2v
```
.
Image-to-Video (i2v): Animate 1–9 reference images into a video using
```
seedance-v2.0-i2v
```
.
Video Extension (extend): Seamlessly continue an existing Seedance 2.0 video using
```
seedance-v2.0-extend
```
.
Multimodal Referencing: Utilize
```
@tag
```
system (
```
@image1
```
,
```
@video1
```
) for style, motion, and rhythm locking.
Audio-Visual Sync: Native high-fidelity sound generation synchronized with visual motion.
Temporal Consistency: Maintain character, clothing, and environment stability across shots.

文本转视频（t2v）：使用
```
seedance-v2.0-t2v
```
，通过导演指令模板生成电影级视频。
图片转视频（i2v）：使用
```
seedance-v2.0-i2v
```
，将1-9张参考图片动作为视频。
视频续拍（extend）：使用
```
seedance-v2.0-extend
```
，无缝续拍已有的Seedance 2.0生成视频。
多模态参考：利用
```
@tag
```
系统（如
```
@image1
```
、
```
@video1
```
）锁定风格、运动节奏与韵律。
音画同步：原生支持高保真音效生成，与视觉运动完美同步。
时间一致性：在多镜头中保持人物、服装与环境的稳定性。

🏗️ Technical Specification: The Director Brief

🏗️ 技术规范：导演指令模板

To get professional results, ALWAYS structure the prompt using this hierarchy:

Component	Instruction Type	Example
Scene	Environment + Lighting	"A rain-soaked cyberpunk street, magenta neon reflections on wet asphalt."
Subject	Identity + Detail	"A woman in a black trenchcoat, determined focus, cinematic skin textures."
Action	Fluid Interaction	"Walking forward through the crowd, coat billowing slightly in the wind."
Camera	Movement + Lens	"Medium tracking shot, 35mm lens, slow dolly backward. Subtle handheld jitter."
Style	Mood + Intent	"Cinematic epic, warm color grade, shallow DOF, rack focus to subject's face."

为获得专业级结果，请务必按照以下层级结构组织提示词：

组成部分	指令类型	示例
场景	环境 + 光线	"被雨水浸湿的赛博朋克街道，洋红色霓虹倒映在潮湿的沥青路面上。"
主体	身份 + 细节	"身穿黑色风衣的女性，眼神坚定，具有电影级皮肤质感。"
动作	流畅交互	"穿过人群向前行走，风衣在风中微微飘动。"
镜头	运动 + 镜头类型	"中景跟拍镜头，35mm焦距，缓慢向后推镜，轻微手持抖动效果。"
风格	氛围 + 意图	"电影级史诗风格，暖色调调色，浅景深，焦点拉至人物面部。"

🧠 Prompt Optimization Protocol

🧠 提示词优化准则

The Agent MUST transform user intent into a technical "Director Brief" before execution.

Technical Grammar: Use camera terms: Dolly In/Out, Crane Shot, Whip Pan, Tracking Shot, Anamorphic Lens, Shallow Depth of Field.
Physics Directives: Use "caustic patterns," "volumetric rays," or "subsurface scattering" instead of "good lighting."
Timecode Notation: For multi-beat scenes, use
```
[00:00-00:05s]
```
format to specify timing.
Tag References: If files provided, use: "Replicate the camera movement of @video1 while maintaining the visual style of @image1."
ORDER MATTERS: Tokens at the start define composition; tokens at the end define texture and micro-motion.
Multi-Image i2v: Provide up to 9 reference images. The model blends aspects (style, identity, environment) across all inputs.

Agent必须先将用户需求转化为专业的「导演指令模板」，再执行生成任务。

专业镜头语法：使用摄影术语：推镜/拉镜、升降镜头、快速摇镜、跟拍镜头、变形镜头、浅景深。
物理指令：使用“焦散图案”、“体积光”或“次表面散射”等术语替代“良好的光线”。
时间码标记：对于多节拍场景，使用
```
[00:00-00:05s]
```
格式指定时间节点。
标签参考：若提供文件，请使用：“复刻@video1的镜头运动，同时保持@image1的视觉风格。”
顺序至关重要：开头的关键词定义画面构图；结尾的关键词定义纹理与微运动。
多图转视频（i2v）：最多可提供9张参考图片，模型会融合所有输入的各方面特征（风格、人物、环境）。

🚀 Protocol: Using Seedance 2

🚀 使用指南：Seedance 2操作流程

Mode 1: Text-to-Video (t2v)

模式1：文本转视频（t2v）

bash

undefined

bash

undefined

Epic reveal shot

bash scripts/generate-seedance.sh
--subject "a hidden temple in the Andes, mist rolling through the canopy"
--intent "epic"
--aspect "16:9"
--duration 10
--quality high
--view

Tense close-up, vertical for social

bash scripts/generate-seedance.sh
--subject "a detective examines a cryptic clue under harsh lamp light"
--intent "tense"
--aspect "9:16"
--duration 5

undefined

bash scripts/generate-seedance.sh
--subject "a detective examines a cryptic clue under harsh lamp light"
--intent "tense"
--aspect "9:16"
--duration 5

undefined

Mode 2: Image-to-Video (i2v)

模式2：图片转视频（i2v）

Animate one or more reference images. Up to 9 images can be supplied — the model synthesizes motion across all of them.

bash

undefined

将一张或多张参考图片动作为视频。最多可提供9张图片——模型会合成所有图片的运动效果。

bash

undefined

Animate a single local image

bash scripts/generate-seedance.sh
--mode i2v
--file hero.jpg
--subject "hero strides forward, coat billowing in slow motion"
--intent "epic"
--aspect "16:9"
--view

Animate from a URL

bash scripts/generate-seedance.sh
--mode i2v
--image "https://example.com/scene.jpg"
--subject "camera slowly pulls back to reveal the full landscape"
--intent "reveal"
--duration 10

Multi-image blending (character + environment + style reference)

bash scripts/generate-seedance.sh
--mode i2v
--file character.jpg
--file environment.jpg
--image "https://example.com/style.jpg"
--subject "character walks through the environment in cinematic style"
--quality high

undefined

undefined

Mode 3: Extend Video

模式3：视频续拍

Continue an existing Seedance 2.0 video seamlessly, preserving visual style, motion, and audio.

bash

undefined

无缝续拍已有的Seedance 2.0生成视频，保留视觉风格、运动节奏与音频。

bash

undefined

Extend with no new prompt (model continues naturally)

bash scripts/generate-seedance.sh
--mode extend
--request-id "abc-123-def-456"
--duration 10

Extend with directional prompt

bash scripts/generate-seedance.sh
--mode extend
--request-id "abc-123-def-456"
--subject "camera continues to pull back, revealing the vast city below"
--intent "reveal"
--duration 10
--quality high
--view

undefined

undefined

Async Pattern (for long jobs)

异步模式（适用于长任务）

bash

undefined

bash

undefined

Submit and get request_id immediately

RESULT=$(bash scripts/generate-seedance.sh --mode i2v --file photo.jpg --async --json) REQUEST_ID=$(echo "$RESULT" | jq -r '.request_id')

Check later

bash ../../../../core/media/generate-video.sh --result "$REQUEST_ID"

---

bash ../../../../core/media/generate-video.sh --result "$REQUEST_ID"

---

⚠️ Constraints & Guardrails

⚠️ 限制条件与规范

No Keyword Soup: DO NOT use "8k, masterpiece, trending." Use technical descriptions: "High-fidelity production grade, 24fps, cinematic grain."
Continuous Action: Describe one fluid motion. Avoid "The man runs and then he stops." Use "The man gradually transitions from a sprint to a sudden stop, chest heaving."
Face Stability: For consistent characters: "Maintain high character consistency, zero facial flicker, persistent clothing details."
Extension Only Works on v2.0:
```
--mode extend
```
requires a
```
request_id
```
from a previous
```
seedance-v2.0-t2v
```
or
```
seedance-v2.0-i2v
```
job.
Aspect Ratios: 16:9, 9:16, 4:3, 3:4 (Seedance 2.0 supports all four).
Duration: 5, 10, or 15 seconds.
Quality:
```
basic
```
(faster) or
```
high
```
(higher fidelity).

避免堆砌关键词：请勿使用“8k、杰作、热门”等词汇，使用专业描述：“高保真制作级画质，24fps，电影级颗粒感。”
动作连贯性：描述一个流畅的动作。避免“男人奔跑然后停下”，应使用“男人从冲刺逐渐过渡到突然停下，胸口起伏。”
面部稳定性：如需保持人物一致性：“保持高度人物一致性，无面部闪烁，服装细节始终不变。”

续拍仅支持v2.0：

--mode extend

需要来自之前

seedance-v2.0-t2v

或

seedance-v2.0-i2v

任务的

request_id

。

宽高比：支持16:9、9:16、4:3、3:4四种宽高比。
时长：支持5秒、10秒或15秒。
画质：
```
basic
```
（更快）或
```
high
```
（更高保真）。

🎭 Prompt Templates (from awesome-seedance community)

🎭 提示词模板（来自awesome-seedance社区）

Cinematic Film Styles

电影风格

[SCENE] Rain-soaked cyberpunk alley, neon signs reflected on wet cobblestones.
[SUBJECT] A lone figure in a weathered trench coat, face obscured by a wide-brim hat.
[ACTION] Walking slowly, each step splashing neon color into the puddles.
[CAMERA] Low-angle tracking shot, anamorphic lens, slow dolly in. Rack focus to face.
[STYLE] Denis Villeneuve aesthetic, high contrast, desaturated blues and magentas. 24fps.

[SCENE] Rain-soaked cyberpunk alley, neon signs reflected on wet cobblestones.
[SUBJECT] A lone figure in a weathered trench coat, face obscured by a wide-brim hat.
[ACTION] Walking slowly, each step splashing neon color into the puddles.
[CAMERA] Low-angle tracking shot, anamorphic lens, slow dolly in. Rack focus to face.
[STYLE] Denis Villeneuve aesthetic, high contrast, desaturated blues and magentas. 24fps.

Advertising / Product Motion

广告/产品动态

[SCENE] Minimalist white studio, single product on a rotating pedestal.
[ACTION] Subtle 360° rotation, product details catching specular highlights.
[CAMERA] Tight medium shot, macro lens pass over surface texture, slow orbit.
[STYLE] Commercial grade, perfect exposure, zero background distraction.

[SCENE] Minimalist white studio, single product on a rotating pedestal.
[ACTION] Subtle 360° rotation, product details catching specular highlights.
[CAMERA] Tight medium shot, macro lens pass over surface texture, slow orbit.
[STYLE] Commercial grade, perfect exposure, zero background distraction.

Action / Physics

动作/物理效果

[SCENE] Desert canyon at sunrise, sandy terrain, long shadows.
[SUBJECT] High-performance sports car accelerating through a turn.
[ACTION] Rear wheels spinning with dust plume, chassis flexing under g-force.
[CAMERA] Low hero angle dolly tracking alongside, then whip pan to lead car.
[STYLE] Hollywood racing film, warm golden grade, motion blur on wheels. 24fps.

[SCENE] Desert canyon at sunrise, sandy terrain, long shadows.
[SUBJECT] High-performance sports car accelerating through a turn.
[ACTION] Rear wheels spinning with dust plume, chassis flexing under g-force.
[CAMERA] Low hero angle dolly tracking alongside, then whip pan to lead car.
[STYLE] Hollywood racing film, warm golden grade, motion blur on wheels. 24fps.

Character Consistency (Martial Arts / Action)

人物一致性（武术/动作）

[SUBJECT] Same fighter throughout: young woman, white gi, black belt, determined expression.
[ACTION] Fluid kata sequence — rising block, stepping side kick, spinning back fist.
[CAMERA] Full-body wide shot, then cut to close-up of fist impact in slow motion.
[STYLE] Maintain identical lighting, clothing, and facial features in every frame. Zero flicker.

[SUBJECT] Same fighter throughout: young woman, white gi, black belt, determined expression.
[ACTION] Fluid kata sequence — rising block, stepping side kick, spinning back fist.
[CAMERA] Full-body wide shot, then cut to close-up of fist impact in slow motion.
[STYLE] Maintain identical lighting, clothing, and facial features in every frame. Zero flicker.

⚙️ Implementation Details

⚙️ 实现细节

Model	Endpoint	Use Case
`seedance-v2.0-t2v`	Text-to-Video	Generate from Director Brief
`seedance-v2.0-i2v`	Image-to-Video	Animate 1–9 reference images
`seedance-v2.0-extend`	Extend Video	Continue a v2.0 generated video

This skill acts as a Cinematographic Wrapper that translates low-level creative intent into high-fidelity technical instructions for the

muapi

core.

模型	接口	使用场景
`seedance-v2.0-t2v`	文本转视频	根据导演指令模板生成视频
`seedance-v2.0-i2v`	图片转视频	将1-9张参考图片动作为视频
`seedance-v2.0-extend`	视频续拍	续拍v2.0生成的视频

该技能作为电影级包装层，将底层创意需求转化为高保真的专业指令，提交给

muapi

核心系统执行。