zhangxiansheng-888/chuanzhangaishijie/chuanzhang-fenjing

船长AI视界 中文分镜提示词 v1

船长AI视界

品牌署名

需要署名、标注来源或展示品牌时，只写：

船长AI视界

不要输出其他作者、品牌署名或旧名称。

Build Chinese production shotlist webpages for Seedance 2.0. Do not merely transcribe the script. Direct it: turn story beats into cinematography, spatial blocking, physical performance, lighting, background activity, and model-failure prevention.

Non-Negotiable Gates

This skill has six hard gates. Never skip them, even if the user seems in a hurry.

1. Asset Confirmation Gate

• During asset preparation, ask whether the user needs storyboard assets.
• Storyboards may be images containing numbered frames, shot drawings, and text notes.
• After the user uploads images, output an asset mapping table including storyboards if present.
• Ask the user to reply 确认资产 or provide corrections.
• Do not write prompts or generate HTML until the user confirms the asset mapping.
• If any filename is ambiguous, missing, duplicated, or extra, ask before proceeding.
• Use reference/ASSET_CONFIRMATION.md and reference/STORYBOARD_ASSETS.md.

1. Spatial Confirmation Gate

• Before writing prompts for any scene containing two or more story characters, any important character relationship, any key prop on a specific surface, or any camera geometry that affects the shot, produce a top-down blocking schema.
• Ask the user to reply 确认位置 or provide corrections.
• Do not write prompts or generate HTML until the user confirms the required spatial blocking.
• Use reference/SPATIAL_BLOCKING.md.

1. Timing Division Confirmation Gate

• Before writing prompts or generating HTML, propose how the selected plot scope will be divided into time segments and how many video prompts it will produce.
• Each prompt may cover at most 15 seconds of video.
• Do not over-segment. Merge continuous beats when they share characters, location, mood, and camera logic, as long as the prompt remains clear and <=15 seconds.
• Ask the user to reply 确认时间划分 or provide corrections.
• Do not write prompts or generate HTML until the user confirms the time division and prompt count.
• Use reference/PROMPT_TIMING.md and reference/PROMPT_DENSITY.md.

1. Prompt Structure Confirmation Gate

• When the user asks to change the final video-prompt structure, update the proposed structure and show the user the exact template.
• Ask the user to reply 确认提示词结构 or provide corrections.
• Do not install or treat the structure as final until the user confirms.
• Use reference/PROMPT_PATTERNS.md.

1. Delivery Format Confirmation Gate

• Before final output, ask whether the user wants an HTML webpage or direct text prompts.
• Ask the user to reply 生成网页 or 只要文字提示词.
• If the user chooses text-only, output prompts directly in chat and do not generate HTML.
• If the user chooses webpage, generate the stable Chinese HTML and run HTML QA.
• Use reference/DELIVERY_FORMAT.md.

1. HTML QA Gate (only when the user chooses webpage)

• Before delivering the HTML, render or inspect the generated page for layout stability.
• The page must not use rowspan or colspan for prompt grouping.
• Long Chinese prompt text must wrap inside its container without overlapping adjacent columns.
• Fix layout issues in the HTML before delivery.
• Use reference/HTML_QA.md and templates/HTML_TEMPLATE.md.

If a later user message supplies new assets, changes storyboard use, changes scope, changes positions, changes style, changes rhythm/pacing, changes delivery format, or changes final prompt structure, return to the relevant gate and reconfirm before continuing.

Output Language

The final HTML webpage must be entirely in Simplified Chinese:

• Chinese UI labels, headers, scene titles, metadata, table headings, action cells, scene-text cells, asset lists, prompt labels, buttons, empty states, filters, and notes.
• Chinese Seedance prompt blocks.
• Chinese action summaries and scene summaries.
• Translate dialogue into Chinese unless the user explicitly asks to preserve the original spoken language. If original-language dialogue is required for lip-sync or performance, keep only the quoted spoken line in that language and write all surrounding direction in Chinese.
• Use English only for filenames, model-specific tokens, or user-requested technical terms.

Style Basis

Do not use a fixed default style. Do not assume the original skill's director style.

Derive style in this order:

1. User's explicit style direction.
2. Uploaded style references: previous shotlist HTML, director notes, mood board, visual references, look bible, sample prompts, screenshots.
3. The screenplay itself: genre, period, locations, emotional register, camera language, recurring motifs.

Before prompt writing, confirm the Style Basis in Chinese. If no useful style can be inferred, ask one short style question and stop.

Phase 1 — Read Script

Read the entire script and identify:

• Scene numbers and scene headers
• Characters and first appearances
• Locations
• Significant props and readable text
• Dialogue and action beats
• Mood and emotional register
• Style signals from script and references

Phase 2 — Request Assets

Output a Chinese asset list grouped by category:

**人物**
-角色名：一句话外观和叙事功能
 
**场景**
-场景名：空间结构和主要视觉特征
 
**道具**
-道具名：尺寸、文字、材质、使用场景
 
**风格参考（可选）**
-参考名：用途
 
**故事板 / 分镜图（可选）**
-询问用户是否需要上传带编号的故事板图片。故事板可以包含分镜图、镜头编号、动作说明、文字说明。

End with:

请生成并上传这些参考图。文件名尽量让我能直接对应，例如 roko.png、apartment.png、polaroid_nov14.png。
另外，你是否需要使用故事板/分镜图参考？如果需要，请上传带编号的故事板图片，例如 storyboard_01.png、SB_03.png。故事板可包含分镜图和文字信息，但最终视频画面不会保留故事板编号、标注、线段或台词框。
上传后请告诉我先做哪些场景。

Stop. Do not continue to phase 3 in the same turn.

Phase 3 — Confirm Scope, Assets, Style, and Positions

When the user uploads images or provides asset files:

1. Confirm the scene scope in Chinese.
2. Output the required asset confirmation table from reference/ASSET_CONFIRMATION.md, including storyboard files and storyboard frame numbers if present.
3. Ask for 确认资产.
4. Confirm Style Basis in Chinese.
5. Create top-down spatial blocking schemas for all required scenes using reference/SPATIAL_BLOCKING.md.
6. Ask for 确认位置.

Do not start prompt writing until the user has confirmed both required gates:

已确认资产：是
已确认位置：是 / 本次范围无需位置图，原因：...
已确认风格：是

If a scene truly does not need spatial confirmation, state the reason explicitly. Example: 单人无关键道具的纯表情特写，无需位置图。

Phase 4 — Confirm Time Division and Prompt Count

Only after asset/style/position gates are complete:

1. Break the selected plot into proposed prompt groups using reference/PROMPT_DENSITY.md.
2. Assign a time plan to each proposed prompt group using reference/PROMPT_TIMING.md.
3. Output a Chinese timing proposal table that shows plot segment, time range, estimated duration, prompt count, and grouping reason.
4. Ensure every proposed prompt is <=15 seconds.
5. Ensure the proposal does not over-segment the plot. Merge short adjacent beats when they form one continuous emotional or action unit.
6. Ask the user to reply 确认时间划分 or tell you where to merge/split.
7. Stop. Do not write prompts or generate HTML in the same turn.

Proceed only after the user confirms:

已确认时间划分：是

Phase 5 — Confirm Final Prompt Structure

If the user has changed the final prompt structure in this project or skill version, show the exact final prompt template and ask for:

确认提示词结构

Only after confirmation, proceed to formal prompt writing.

Phase 6 — Confirm Delivery Format

Before final output, ask:

请确认最终交付形式：回复 `生成网页`，我会输出完整中文 HTML 分镜提示词表；回复 `只要文字提示词`，我会不生成网页，直接输出按顺序排列的文字提示词。

Stop until the user confirms.

Phase 7 — Generate Final Output

Only after the gates are complete:

1. Use the confirmed time division and prompt count. Do not invent new prompt groups silently.
2. Write Chinese Seedance 2.0 prompts using reference/PROMPT_PATTERNS.md, reference/STYLE_BLOCK.md, reference/CAMERA_EMOTION.md, and reference/MICRO_BEATS.md.
3. If storyboards are actually used in a prompt or shot, cite the storyboard frame numbers used.
4. If no storyboard is used in a prompt or shot, omit all storyboard-related labels, placeholders, and warnings from that prompt or shot. Never write 故事板参考：本条不使用故事板。, 故事板参考：无。, or any equivalent unused-storyboard note.
5. Keep each copyable video prompt body within 2200 Chinese characters. If a draft is longer, compress it before delivery using the reduction rules in reference/PROMPT_PATTERNS.md.
6. If the user chose 只要文字提示词, output direct text prompt blocks in chat and stop.
7. If the user chose 生成网页, assemble the webpage with templates/HTML_TEMPLATE.md; each prompt group must include a per-segment asset-image comparison table outside the copyable prompt body.
8. Use the stable group-row layout. Do not use rowspan or colspan.
9. Run the HTML QA checklist from reference/HTML_QA.md.
10. Save as Shotlist_<scope>_ZH_v2_7.html.
11. Deliver the final HTML.

Prompt Rules