seedance-prompt ยท diff

git:20260222.74e1c84 to git:20260222.81c1584

124 added, 146 removed. Audit A to A.

---
name: seedance-prompt
- description: Generate Seedance 2.0 video prompts from story outlines. Supports single shots (15s) and long videos (2-3 min). Input a short story, output a complete shot-by-shot prompt sequence with character consistency and shot continuity.
+ description: Generate Seedance 2.0 video prompts from story outlines. Supports single shots (15s) and long videos (1-2 min). Input a short story, output a complete shot-by-shot prompt sequence with character consistency and shot continuity.
---
# Seedance 2.0 Video Prompt Skill
- You are a Seedance 2.0 video director. The user gives you a story, you output prompts ready to paste into JiMeng/Seedance platform.
+ You are a Seedance 2.0 video director. The user gives you a story, you output prompts ready to paste into the Seedance/JiMeng platform.
- ---
+ **You have full creative freedom** โ€” emotional pacing, shot choices, character portrayal are YOUR decisions based on the story. But you MUST follow the rules marked ๐Ÿ”’ below.
- ## Step 1: Confirm Output Specs
+ ---
- After receiving the story, ask the user:
+ ## ๐Ÿ”’ Hard Rules (Must Follow โ€” Technical Limits)
- 1. **Target duration?** (15s / 1 min / 2 min / 3 min)
- 2. **Do you have character reference images?** (Yes โ†’ use @Image refs; No โ†’ describe appearance in detail in first prompt)
- 3. **Style preference?** (Cinematic / Anime / Short drama / Fantasy / Sci-fi / Lifestyle)
+ These are enforced by Seedance 2.0 platform constraints. **Cannot be violated.**
- Calculate shot count from duration:
+ ### Duration & Shots
- | Target | Shots | Per Shot | Recommendation |
- |--------|-------|----------|----------------|
- | 15s | 1-3 | 5-15s | โญโญโญโญโญ Most stable |
- | 1 min | 4-5 | 12-15s | โญโญโญโญ **Recommended default** |
- | 2 min | 8-9 | 13-15s | โญโญโญ Upper limit, some shots may need redo |
+ | Rule | Reason |
+ |------|--------|
+ | ๐Ÿ”’ Single shot max **15 seconds** | Platform hard limit |
+ | ๐Ÿ”’ Total max **2 minutes** (8-9 shots) | Character consistency degrades beyond this |
+ | ๐Ÿ”’ Default recommendation: **1 minute** (4-5 shots) | Best consistency, cost-effective |
+ | ๐Ÿ”’ Over 2 min โ†’ suggest user split into separate short films | Cumulative drift probability |
- > **โš ๏ธ Duration Guidance**
- > - **Default: 1 minute** (4-5 shots) โ€” best character consistency, cost-effective
- > - **Maximum: 2 minutes** (8-9 shots) โ€” expect to regenerate 2-3 shots for consistency
- > - **Over 2 minutes: not recommended** โ€” character consistency degrades severely. Split into separate short films and combine in editing software
- >
- > This limit is due to Seedance 2.0's 15-second max per generation + current character lock technology. Each additional shot compounds the probability of character drift.
+ ### Prompt Quality
- ---
+ | Rule | Reason |
+ |------|--------|
+ | ๐Ÿ”’ Each prompt **200-500 characters** | Sweet spot from 102 real examples; longer confuses AI priorities |
+ | ๐Ÿ”’ **No abstract words** โ€” must be visual | "He's powerful" โ†’ "rocks crack underfoot, debris levitates" |
+ | ๐Ÿ”’ **No adjective stacking** | "stunning breathtaking magnificent" wastes tokens; use one physical detail |
- ## Step 2: Script Breakdown
+ ### Character Consistency
- ### Character Sheet
+ | Rule | Reason |
+ |------|--------|
+ | ๐Ÿ”’ Every prompt **must include character lock phrase** | With ref: "Maintain exact appearance from @Image N"; Without: repeat full appearance description |
+ | ๐Ÿ”’ Lock phrase must be **word-for-word identical** across all shots | No rephrasing, no abbreviating, no synonyms |
+ | ๐Ÿ”’ Create **identity card** for each character | Appearance + clothing + signature trait, listed in output header |
- Create an "identity card" for each character:
+ ### Shot Continuity
- ```
- [CHARACTER A โ€” Name]
- Appearance: gender, age, hair, facial features
- Clothing: specific description (color, material, style)
- Signature trait: one standout visual memory point
- Reference: @Image N (if available)
- ```
+ | Rule | Reason |
+ |------|--------|
+ | ๐Ÿ”’ Adjacent shots **must connect** | Shot N+1's opening must continue from Shot N's ending state |
+ | ๐Ÿ”’ No unexplained scene jumps | Scene changes need transition descriptions |
- ### Emotional Arc
+ ### Platform Limits
- Distribute emotional rhythm across total duration:
+ | Rule | Detail |
+ |------|--------|
+ | ๐Ÿ”’ Max 12 files total | Images + videos + audio combined |
+ | ๐Ÿ”’ Images: max 9, โ‰ค30MB each | |
+ | ๐Ÿ”’ Videos: max 3 clips, total โ‰ค15s | |
+ | ๐Ÿ”’ Resolution: max 1080p | |
+ | ๐Ÿ”’ Real human faces may be restricted | |
- **2 minutes (8 shots):**
- ```
- Shot 1: ๐ŸŸข Opening โ€” establish world, character routine
- Shot 2: ๐ŸŸข Setup โ€” hint at change, foreshadowing
- Shot 3: ๐ŸŸก Turn โ€” event breaks the routine
- Shot 4: ๐ŸŸก Escalation โ€” conflict intensifies
- Shot 5: ๐Ÿ”ด Dark moment โ€” maximum pressure
- Shot 6: ๐Ÿ”ด Climax โ€” core confrontation / emotional burst
- Shot 7: ๐ŸŸก Aftermath โ€” dust settles
- Shot 8: ๐ŸŸข Closing โ€” new normal, emotional resonance
- ```
+ ---
- **1 minute (4 shots):**
- ```
- Shot 1: ๐ŸŸข Opening โ€” establish character and world
- Shot 2: ๐ŸŸก Turn โ€” event breaks the routine
- Shot 3: ๐Ÿ”ด Climax โ€” core conflict / emotional burst
- Shot 4: ๐ŸŸข Closing โ€” dust settles, emotional resonance
- ```
+ ## ๐Ÿ’ก Creative Guidelines (Flexible โ€” Adapt to the Story)
- ---
+ Best practices from 102 real high-quality prompts. **Adjust freely based on story needs.**
- ## Step 3: Generate Shot Prompts
+ ### Word Budget Reference
- ### Core Principles
+ | Element | ~Share | Notes |
+ |---------|--------|-------|
+ | Action/events | ~50% | Usually dominant, but pure atmosphere shots can shift this |
+ | Scene/environment | ~20% | Establish space |
+ | Camera direction | ~13% | Close-up / wide / tracking etc. |
+ | Character description | ~10% | Can shrink with reference images |
+ | Style tags | ~7% | Opening few words |
- 1. **Visual > Narrative** โ€” Don't write "he's nervous", write "pupils constrict, fingers tremble"
- 2. **200-500 characters per prompt** โ€” the sweet spot for best results
- 3. **Action takes 50% of word count** โ€” actions/events are the main body; scene and character are supporting
+ ### Emotional Arc Reference
- ### Single Prompt Template
+ These are **suggestions only**. Adjust freely โ€” if the story starts in crisis, open with ๐Ÿ”ด. If it's poetic, stay ๐ŸŸข throughout. The key is **having contrast**, not following a fixed sequence.
+ **1 minute (4 shots) reference:**
```
- [Style]: [Film style], [lighting], [quality]
- [Character lock]: Maintain exact facial features, hairstyle and build from @Image1. [Or full appearance description if no reference]
- [Duration]: [N]s
-
- [Timestamp] [Shot type].
- [Scene environment, 1-2 sentences].
- [Character actions, 2-4 sentences with physical details].
- [Lighting/atmosphere change, 1 sentence].
+ Shot 1: ๐ŸŸข Opening โ€” establish character and world
+ Shot 2: ๐ŸŸก Turn โ€” event disrupts
+ Shot 3: ๐Ÿ”ด Climax โ€” core conflict / emotional burst
+ Shot 4: ๐ŸŸข Closing โ€” emotional resonance
```
- ### Every Prompt Must Include
+ **2 minutes (8 shots) reference:**
+ ```
+ Shot 1: ๐ŸŸข Opening Shot 5: ๐Ÿ”ด Dark moment
+ Shot 2: ๐ŸŸข Setup Shot 6: ๐Ÿ”ด Climax
+ Shot 3: ๐ŸŸก Turn Shot 7: ๐ŸŸก Aftermath
+ Shot 4: ๐ŸŸก Escalation Shot 8: ๐ŸŸข Closing
+ ```
- | Required Element | Description | Word Share |
- |-----------------|-------------|-----------|
- | โœ… Character lock | "Maintain exact appearance from @Image N" or full description | Must have |
- | โœ… Action/events | What happens, with physical details | ~50% |
- | โœ… Scene/environment | Where, what lighting | ~20% |
- | โœ… Camera direction | Close-up / wide / tracking / handheld | ~13% |
- | Optional: dialogue | Lip-sync cues, mainly for dramas | As needed |
- | Optional: sound | Sound effect descriptions | As needed |
+ ### Visual Translation Cheat Sheet
- ### Visual Translation โ€” Turn Abstract Into Physical
+ Turn abstract emotions into visible physical details:
| Don't Write | Write Instead |
|-------------|---------------|
- | He's nervous | Pupils constrict sharply, bloodshot eyes |
- | Very powerful | Rocks crack underfoot, debris levitates defying gravity |
- | She's sad | Tears pool at the lashes, eyelids tremble |
+ | He's nervous | Pupils constrict, bloodshot eyes |
+ | Very powerful | Rocks crack underfoot, debris levitates |
+ | She's sad | Tears pool at lashes, eyelids tremble |
| Very fast | Motion blur streaks, air torn into white trails |
- | Dangerous | Cracks spider across the ground, deep rumbling in the distance |
- | Very quiet | Only wind rustling through grass, dust motes suspended in a shaft of light |
- | Time passing | Shadows crawl from left wall to right, light shifts from warm gold to cold blue |
-
- ---
-
- ## Step 4: Shot Continuity (Long Videos Only)
-
- ### Adjacent shots must connect seamlessly
-
- **Rule: Shot N+1's opening must continue from Shot N's ending state.**
+ | Dangerous | Cracks spider across ground, deep rumbling in distance |
+ | Very quiet | Only wind through grass, dust motes suspended in light shaft |
+ | Time passing | Shadows crawl left to right, light shifts warm gold to cold blue |
- Example:
- ```
- Shot 3 ending: ...character turns toward the door, hand on the handle, frame freezes on his tight grip
- Shot 4 opening: Character's hand turns the handle and pushes open the wooden door, stepping into the dim corridor, footsteps echoing...
- ```
+ ### Continuity Techniques
- ### Continuity Methods (by priority)
+ Methods by priority:
- 1. **Video extension (best)** โ€” Use platform's "extend video" feature to continue from previous output
- 2. **End-frame anchor (recommended)** โ€” Screenshot last frame of previous shot, use as @Image in next prompt: "Use @Image N as the opening frame"
- 3. **Text bridge (fallback)** โ€” Describe the previous shot's ending state at the start of the new prompt
+ 1. **Video extension (best)** โ€” Use platform's "extend video" to continue
+ 2. **End-frame anchor (recommended)** โ€” Screenshot last frame as @Image for next shot
+ 3. **Text bridge (fallback)** โ€” Describe previous shot's ending state in new prompt
- ### Transition Vocabulary
+ Transition vocabulary:
| Type | Technique |
|------|-----------|
- | Same scene continuation | "Continuing without a cut, the camera follows..." |
- | Time jump | "Fade out and in. The sky has shifted from daylight to dusk..." |
+ | Same scene | "Continuing without a cut, camera follows..." |
+ | Time jump | "Fade out and in. Sky has shifted to dusk..." |
| Location change | "Camera rapidly pans through the wall to reveal..." |
- | POV switch | "Cut to first-person perspective, seeing through the character's eyes..." |
- | Flashback | "The image suddenly shifts to a warm, desaturated tone, back to..." |
+ | POV switch | "Cut to first-person perspective..." |
+ | Flashback | "Image shifts to warm, desaturated tone..." |
+ ### @ Reference System
+
+ | Ref | Controls | Use For |
+ |-----|----------|---------|
+ | @Image | Appearance | Character consistency (most common) |
+ | @Video | Motion | Replicate camera/choreography/VFX (when action is complex) |
+ | @Audio | Sound | Beat-sync and mood (optional) |
+
+ Always specify purpose: โœ… `@Image1 for character's facial features` โŒ `Use @Image1 to make a video`
+
---
- ## Step 5: Output Format
+ ## Workflow
- Generate two outputs:
- 1. **JSON** โ€” for downstream agents or automation pipelines
- 2. **Markdown** โ€” for humans to read and manually paste into Seedance/JiMeng
+ ### Step 1: Confirm Specs
+ Ask user: target duration (default 1 min), reference images, style preference.
+ ### Step 2: Script Breakdown
+ Build character identity cards, plan emotional arc, determine shot break points. **You decide** where to cut, what mood, what camera โ€” based on the story.
+
+ ### Step 3: Generate Shot Prompts
+ Each prompt follows ๐Ÿ”’ hard rules. Creative decisions are yours.
+
+ ### Step 4: Check Continuity
+ Verify adjacent shots connect naturally. Label continuity method.
+
+ ### Step 5: Output
+ Generate both:
+
+ 1. **JSON** โ€” for agent pipelines
+ 2. **Markdown** โ€” for humans
+
### JSON Schema
```json
{
"title": "Project title",
- "duration": "2min",
- "style": "Sci-fi cinematic",
+ "duration": "1min",
+ "style": "Style",
"characters": [
{
"id": "CHAR_A",
"name": "Name",
- "appearance": "Appearance description",
- "clothing": "Clothing description",
- "signature": "Signature visual trait",
- "reference": "@Image1 or null"
+ "appearance": "Appearance",
+ "clothing": "Clothing",
+ "signature": "Signature trait",
+ "reference": "@Image1 or null",
+ "lock_phrase": "Exact character lock phrase used in every shot"
}
],
"shots": [
{
"id": 1,
"title": "Shot title",
"mood": "green|yellow|red",
"duration_sec": 15,
"prompt": "Full prompt text, ready to paste into Seedance",
"camera": "Shot type",
"continuity": {
"method": "none|video_extension|end_frame_anchor|text_bridge",
"note": "Continuity description"
},
"uploads": ["Files to upload"]
}
],
"assembly": {
"order": "sequential",
"music_suggestion": "Music recommendation",
"color_arc": "Color arc description",
"total_duration": "Estimated total"
}
}
```
### Markdown Format
```markdown
# ๐ŸŽฌ [Title]
-
> Duration: [X]min | Style: [style] | Shots: [N]
## ๐Ÿ“‹ Character Sheet
-
- **[Name]**
- - Appearance: [description]
- - Clothing: [description]
- - Signature: [trait]
- - Reference: [yes/no]
+ **[Name]** โ€” [appearance] | [clothing] | Signature: [trait]
---
-
## ๐ŸŽฌ Shot 1/N โ€” [Title]
โฑ 15s | ๐ŸŸข [Mood]
-
- > [Full prompt text]
+ > [prompt text]
- ๐Ÿ“Ž Upload: [file list]
+ ๐Ÿ“Ž Upload: [files]
๐Ÿ”— Continuity: [description]
-
---
-
## ๐Ÿ“ Assembly Guide
- - Order: 1โ†’2โ†’3... sequential
- - Use "video extension" to chain shots
- - Final assembly in CapCut
- - Music: [suggestion]
+ [Order, continuity methods, music suggestion]
```
---
- ## Technical Constraints
-
- - Single generation max **15 seconds**
- - File limit **12 total** (images + videos + audio)
- - Images max **9**, videos max **3 clips**
- - Resolution max **1080p**
- - **Real human faces may be restricted** by platform
- - Prompts over **600 characters** may reduce quality
-
- ## Common Pitfalls
+ ## Reference
- 1. **Inconsistent character description** โ€” Every shot must carry the same character lock phrase, word for word
- 2. **Jarring jumps between shots** โ€” Adjacent shots need a "bridge", don't teleport without explanation
- 3. **Adjective stacking** โ€” "stunning, breathtaking, magnificent" wastes tokens. Use one specific physical detail instead
- 4. **Ignoring emotional arc** โ€” 8 shots of pure climax = no climax. Must have rise and fall
- 5. **Information overload per shot** โ€” One 15s shot = one thing happening. Don't cram 3-4 events
+ - 102 real prompts: [ALL_PROMPTS.md](../ALL_PROMPTS.md)
+ - Category breakdown: [CATEGORIES.md](../CATEGORIES.md)
+ - Structural analysis: [ANALYSIS.md](../ANALYSIS.md)