honeybee1130

Seedance

內容創作v1.0.0
作者:honeybee1130更新於 2026-05-11seedance@clawhub_honeybee1130/seedance
短影片指令碼#Video Generation#Prompt

據 AlphaMoat 統計,截至 2026-10,Seedance 累計下載 10,088 次、獲 4 個星標,按累計下載量在本次參與排名的 10,914 個 AI 技能中排名第 2,702 位,在「內容創作」分類(927 個)中排名第 273 位。

列表預設按累計下載量排序,也可按星標排序;詳情頁排名按累計下載量計算。

累計下載
1萬
星標
4
全站下載量排名
#2,702
全站前 25% · 「內容創作」前 30%
平台累計下載星標評論
ClawHub ↗33313-
SkillHub ↗1萬4—

各平台的計數口徑不同,統計時間不一致。

這個技能是做什麼的

按主體、動作、鏡頭、風格、音訊和約束組織,適用時使用參考標籤。提供的是提示模板,不自行生成影片。使用前審閱提示和源文相關段落,移除偽裝受限 IP 或規避稽核的指導,遵守目標平台規則。

資料釋出於 2026-10-27

可以用來做什麼

  • 起草結構化影片提示詞

什麼時候用

  • 明確需要為 Seedance 編寫提示詞時

示例提示詞

以下是根據技能文件生成的示例,不代表實際執行結果。

為原創場景編寫含鏡頭、音訊與一致性約束的提示詞,避免偽裝受限IP,並按平台規則審閱。

月度資料趨勢

累計下載1萬
69857411.1萬2026-03 · 累計下載 13942026-05 · 累計下載 22532026-06 · 累計下載 27662026-07 · 累計下載 35372026-08 · 累計下載 42292026-09 · 累計下載 54622026-10 · 累計下載 1萬26-0326-0926-10
星標4
1342026-03 · 星標 12026-05 · 星標 32026-06 · 星標 42026-07 · 星標 42026-08 · 星標 42026-09 · 星標 42026-10 · 星標 426-0326-0926-10

2026-09 起統計方法調整:同名技能按作者分別統計,並覆蓋全部技能。圖中虛線為調整前的月份,與之後的資料不直接比較,也不計算跨越調整的增長。 該技能有同名的其他版本,調整前的資料歸屬當前下載量最大的這一個。

月份累計下載環比星標
2026-101萬—4
2026-095462—4
2026-084229+19.6%4
2026-073537+27.9%4
2026-062766+22.8%4
2026-052253—3
2026-031394—1

SKILL.md 原文

以下是作者釋出的原始檔案,按純文字顯示,本站未做修改或稽核;其中的連結不可點選,命令請核實後再執行。抓取於 2026-10-23。

11 KB · ClawHub v1.0.0
---
name: seedance
description: Generate cinematic video prompts for Seedance 2.0 (ByteDance). Use when user asks for video prompts, Seedance prompts, AI video generation prompts, or anything related to Seedance 2.0 prompt crafting. Triggers on: seedance, video prompt, AI video, text to video prompt, seedance prompt.
---

# Seedance 2.0 Prompt Generator

Generate production-ready prompts for ByteDance's Seedance 2.0 AI video model.

## Prompt Architecture
展開全部(共 220 行)

Every prompt follows this strict order. Deviating causes drift.

```
Subject → Action → Camera → Style → Audio → Constraints
```

### 1. Subject (WHO/WHAT)
- One primary subject. Multiple subjects split model attention.
- Include: age/material, clothing, distinguishing features
- Example: "Wooden Koda creature with glowing orange eyes, green gems on head, purple cape"

### 2. Action (WHAT HAPPENS)
- Specific verb phrases, present tense
- Describe beat by beat for complex sequences
- One action per beat. Chain beats chronologically.
- Example: "walks to cliff edge, pauses, turns head slowly to camera, cape billowing"

### 3. Camera (HOW WE SEE IT)
- Shot size FIRST: wide / medium / close-up / extreme close-up
- Movement SECOND: dolly-in, dolly-out, track left/right, crane up/down, pan, tilt, handheld, gimbal, locked-off
- Angle: eye level, low angle, high angle, bird's eye, Dutch
- Lens feel: wide (24-28mm), normal (35-50mm), telephoto (85mm+)
- ONE verb per shot. Compound moves = separate beats: "Start: slow dolly-in. Then: gentle pan right for final 2s"

**Shot Cheat Sheet:**
| Shot | Use | Pair With |
|------|-----|-----------|
| Wide | Establish space/context | Slow dolly, locked-off |
| Medium | Subject + context, dialogue | Handheld (personal), gimbal (polished) |
| Close-up | Detail, emotion | Tiny push-in, avoid pans |
| Tracking | Movement, energy | Lateral follow, side profile |

### 4. Style (THE LOOK)
- ONE visual anchor > six adjectives
- Lighting: key light type, time of day, practical sources
- Color treatment: muted/saturated/monochrome/specific palette
- Texture: film grain, clean digital, anamorphic, etc.
- Reference format: "[film/artist/era] aesthetic"

### 5. Audio (WHAT WE HEAR)
- Seedance 2.0 generates dual-channel stereo audio
- Specify: background music genre, environmental sounds, dialogue/VO, silence
- Audio syncs to visual action automatically
- Example: "ambient wind, distant thunder, no music, footsteps on stone"

### 6. Constraints (GUARDRAILS)
- Ban list: no text overlays, no extra characters, no snap zooms, no watermarks
- Timing: hold frames, beat durations, total length (5s or 10s for testing, 15s max)
- Consistency: "maintain character identity throughout", "no morphing"
- Physics: "realistic cloth physics", "gravity-accurate"

## Reference System (@Tags)

When user provides images/videos, use @tags:
- `@Image1`, `@Image2`, etc. for uploaded images
- `@Video1`, `@Video2`, etc. for uploaded videos
- `@Audio1`, etc. for uploaded audio

**Usage patterns:**
- Character identity: "@Image1 is the main character"
- First/last frame: "@Image1 as first frame, @Image2 as last frame"
- Motion transfer: "@Image1 performs the dance from @Video1"
- Style reference: "match the color palette of @Image3"
- Multi-reference: up to 9 images + 3 videos + 3 audio clips

## Prompt Templates

### Cinematic Scene
```
[Scene type] style. [Subject with details]. [Action beat 1], [action beat 2], [action beat 3].
[Camera: shot size], [movement], [angle], [lens feel].
[Lighting description], [color treatment], [texture/grain].
[Audio: music/sfx/ambience].
[Constraints: bans, timing, consistency notes].
```

### Multi-Shot Narrative
```
Shot 1: [Wide/establishing]. [Scene description]. [Camera movement]. [Duration].
Shot 2: [Medium/close]. [Character action]. [Camera movement]. [Duration].
Shot 3: [Close-up/detail]. [Emotional beat]. [Camera movement]. [Duration].
[Overall style], [color grade], [audio design].
[Constraints].
```

### Action Sequence
```
[Genre] action sequence. [Setup description].
Beat 1: [Action], [camera follows with movement type].
Beat 2: [Reaction/counter], [cut to shot size], [slow motion if needed].
Beat 3: [Resolution], [camera pulls to reveal].
[Style: reference film/show]. [Audio: impact sounds, score].
[Constraints: physics accuracy, no artifacts].
```

## Negative Prompt Checklist (pick 3-5 per generation)

**Visual noise:** no text overlays, no watermarks, no floating UI, no lens flares
**Identity drift:** no extra characters, no crowd, no mirrors reflecting others
**Camera chaos:** no snap zooms, no whip pans, no Dutch angles, no jump cuts
**Body artifacts:** no extra fingers, no deformed hands, no warped objects, no melting edges
**Branding:** no logos, no labels, no recognizable brands
**Color:** no neon lighting, no heavy teal/orange, no cartoon saturation
**Environment:** no rain/fog/smoke unless stated, no confetti, no dust particles

## Advanced: Clean High-Motion Technique

Learned from real-world results. These techniques produce sharp, blur-free motion even at extreme speed.

### The Continuous Shot Lock
- Declare "single continuous shot" upfront — forces temporal coherence, prevents inter-scene interpolation artifacts
- The model treats the entire generation as one fluid motion path instead of stitched segments

### Physics-Motivated Camera
- Every camera move needs a VERB with physical motivation: dive, slingshot, whip, dart, blast
- Never say "dynamic camera" — say WHY the camera moves (following subject, reacting to explosion, releasing into reveal)
- Camera attached to subject ("lock-on," "staying glued") = subject stays sharp because relative motion is zero

### Environmental Anchoring
- Scatter static reference geometry throughout: walls, arches, furniture, hanging objects
- The model needs stable background to render motion AGAINST — parallax creates perceived speed without subject blur
- Static objects streaking past a centered subject = clean speed

### Scale Progression Arc
- Structure as Macro → Micro → Macro (wide establish → tight detail → wide reveal)
- Gives model clear resolution targets at each stage — doesn't try to render everything at once
- The "reveal" at the end (pulling wide after sustained close action) creates cinematic payoff

### Sensory Render Instructions (Not Mood Words)
- Replace adjectives with computable effects: "heat haze" not "hot," "grit snapping off ledge" not "dusty," "mist turning into rainbow" not "magical"
- Each detail should be something the model can physically simulate

### Rhythm Through Verbs
- Pacing lives in action chain length, not "hold for Xs" timers
- Quick beat: "snaps a last-inch swerve" (short clause = fast)
- Sustained beat: "threads through hanging laundry lines and open windows in one fluid line" (long clause = flowing)
- Climax: contrast — "sudden calm" after chaos = tension release

### Reference Prompt (Proven Clean High-Motion)
```
Speeder chase across a cliff city (single continuous shot)
From a monumental cliffside city carved into stone, the camera dives toward a tiny streak of light ripping along a narrow ledge-road. Lock-on: a speeder hugging the wall at insane speed. The camera slingshots ahead, whips back, then drops tight to the rear thrusters: heat haze, grit snapping off the ledge, warning lights flashing. A collapsing balcony rains debris; the rider snaps a last-inch swerve under a falling arch, then threads through hanging laundry lines and open windows in one fluid line. The camera darts through the same openings, staying glued to the motion. One final bend and sudden calm: the camera blasts outward into a reveal of the city opening onto a boundless waterfall-fed valley, mist turning into rainbow.
```

## Pro Tips

1. **High-res references** — 2K/4K input images = better output. Blurry in = blurry out
2. **Test at 5s first** — iterate fast, extend to 10-15s once the motion is right
3. **One change at a time** — don't rewrite the whole prompt on a miss, tweak one element
4. **Creativity/Consistency sliders** — 60% consistency / 40% creativity is the sweet spot
5. **Beat timing** — write "hold for 2s" or "pause 1s" to control pacing
6. **Compound camera = separate beats** — never jam two movements in one clause
7. **Style anchor > adjective soup** — "Blade Runner 2049 aesthetic" > "cinematic dark moody neon futuristic"
8. **Environmental audio** — even without dialogue, specify ambient sounds for immersion

## ⚠️ Content Policy Rules (CRITICAL)

Seedance is a **video diffusion model with no internet access**. It does not know:
- "Otherside" — means nothing to it
- "BAYC," "Koda," "Other Games" — proprietary names get flagged
- Gaming/metaverse/NFT terminology — triggers content moderation
- Crowd scenes — flagged automatically
- Any brand, game title, or IP name

**The rule: Never use brand names. Describe what you see, not what it's called.**

### Translation Layer — Concept → Visual Description

| What you mean | What to write in the prompt |
|---|---|
| Otherside world | "alien landscape with bioluminescent terrain, floating rock formations, purple and teal crystal growths, twin moons visible through violet sky" |
| Otherside biome | "ancient stone ruins overtaken by glowing fungal growth, rivers of liquid light flowing through cracked earth" |
| Otherside portal | "swirling circular energy vortex of purple and gold light, crackling at the edges, suspended in midair" |
| Koda character | "small wooden creature with bark-textured skin, glowing amber eyes in a dark hollow face, green teardrop gems across the top of its head, dark feathered collar, flowing purple cape, small dark armored clawed hands, chibi proportions" |
| Other Games event | "large gathering of stylized 3D avatar characters in a glowing arena, spotlights, festive atmosphere" (avoid "crowd" — use "scattered figures" or "a handful of characters") |
| Gaming/metaverse | Use: "virtual world," "digital realm," "fantastical landscape," "animated environment" |

### Banned Words (will trigger rejection)
- Otherside, Yuga, BAYC, Koda, ApeCoin, NFT, metaverse, web3, blockchain, cryptocurrency
- crowd, mob, large group, mass of people
- Any real brand name, game title, or IP

## Koda-Specific Prompts (Other Games IP)

When generating Koda content:
- **NEVER say "Koda"** — describe the character visually every time
- Use: "small wooden creature with bark-textured skin, glowing amber eyes in a dark hollow face, green teardrop gems across the top of its head, dark feathered collar, flowing purple cape, small dark armored clawed hands, chibi proportions"
- **NEVER say "Otherside"** — describe the environment visually
- Use: "alien landscape with bioluminescent terrain, floating rock formations, purple and teal crystal growths"
- Maintain character consistency across shots
- Always provide Honey B's Koda image as @Image1 for I2V generations (best result)
- Test at 5s first to confirm character renders correctly before extending

## Platform Access
- Jimeng AI (即梦): jimeng.jianying.com → Video Generation → Seedance 2.0
- Doubao App: dialogue box → Seedance 2.0 → select 2.0 model
- Volcano Engine: experience center → Doubao-Seedance-2.0

## When User Asks for a Prompt

1. Ask what scene/concept they want (or use their description)
2. Determine: T2V (text only), I2V (image + text), or R2V (multi-reference)
3. Pick the right template
4. Fill in all 6 layers (subject → constraints)
5. Add 3-5 relevant negative constraints
6. Output the final prompt ready to paste
7. Suggest aspect ratio (16:9 cinematic, 9:16 social, 1:1 square)
8. Suggest starting duration (5s test → extend)

「內容創作」分類下的熱門技能

李白.Skill內容創作

李白.Skill:潤色專家 檢測並消除中文文字中的AI生成痕跡,將生硬的機器輸出轉化為自然流暢的人類文筆。融合規則檢測、分層改寫(詞彙→句式→內容→邏輯→語氣)和六步質控流水線,確保最終文字透過人工審查。

↓ 125萬★ 121user_741dc82b
libai-skill內容創作

李白.Skill:潤色專家 檢測並消除中文文字中的AI生成痕跡,將生硬的機器輸出轉化為自然流暢的人類文筆。融合規則檢測、分層改寫和質控流水線,確保最終文字透過人工審查。

↓ 101.8萬★ 11user_814dbe54
文章去AI味工具內容創作

識別並去掉文字里的 AI 寫作痕跡,讓文字讀起來更像人寫。

↓ 95.1萬★ 767user_ab5ae6ee
影片號爆款短影片拆解(付費版:全能)【零一數科·出品】內容創作

【零一數科·出品】影片號爆款短影片拆解(全能版)(付費版)。影片號爆款短影片拆解(付費版:全能)。一鍵拆解爆款,把一條影片號影片拆成結構分段、爆款歸因、六維評分和可借鑑策略,照著學照著抄。支援影片號分享連結和本地影片,付費全能版遠端分析,1–5 分鐘出視覺化 HTML 報告。

↓ 91.9萬★ 116u_c19e970c
影片號掛車爆單短影片指令碼生成【零一數科·出品】內容創作

【零一數科·出品】影片號掛車爆單短影片指令碼生成(付費版)。一條連結二創指令碼,丟進爆款影片號連結即以此骨架改寫成你的掛車指令碼,或按方向/目的/商品原創多條口播指令碼(含分段表、口播全文、六維質檢評分)。

↓ 55.4萬★ 39u_c19e970c
Humanizer內容創作

把文字里 AI 寫作的典型痕跡挑出來並改寫得更自然,同時保留原意的寫作編輯技能。

↓ 50.9萬★ 1116B*T
中文公文寫作內容創作

用於中文公文、事務性材料和新聞稿件的起草、改寫、壓縮、潤色、審校、文種核對、去口語化、降 AI 味及 Word 格式處理,適用於機關、企事業單位、學校和新聞機構。涵蓋申請、請示、報告、通知、通告、意見、決定、決議、議案、公報、命令、函、覆函、批覆、說明、方案、紀要、公告、公示、通報、制度、規定、辦法、細則、操作規程、工作要點、總結、調研、講話、致辭、主持詞、述職、可研、審查材料、技術需求、新聞訊息、編者按、新聞評論,以及採購、整改、反饋和 AI 算力等場景。

↓ 41.3萬★ 84gongyu0918
影片號爆款文案生成(體驗版)【零一數科·出品】內容創作

【零一數科·出品】影片號爆款文案生成(體驗版)。免費免登入出指令碼,給創作目的、行業、受眾、選題就按爆款結構(Hook-中段-CTA、節奏標籤)產出口播指令碼,資訊不全也能按爆款潛質兜底。本地執行、無需 Key,零成本上手。

↓ 33.9萬★ 94u_c19e970c

資料月份:2026-10 · 下載/星標數聚合自公開技能註冊庫(ClawHub、SkillHub),按月更新。口徑說明見資料方法。