YGtemple/video-skills-collection

AI Short Drama Master Suite

🏷️ Skill Classification

Category: [Live-Action Oriented] (Hybrid Skill – Primary category is live-action realistic short drama, includes stylized/comic drama branch)

• 🎬 Typical Task Scenarios: ["Help me create a live-action realistic short drama from scratch (90 seconds or 5 minutes)"; "Break this novel/script into storyboards and batch generate videos"; "Replicate a short drama following a reference film (all assets remade, shot-by-shot alignment)"; "Produce → Review → Consistency → Hard-Cut Assembly → Multi-Platform Delivery (step-by-step with options, waiting for confirmation)"]
• 🚫 Non-Applicable Scenarios: [Pure editing/compositing/media asset processing/extraction → use jianying-editor, byted-mediakit series, doubao-video-extract; Commercial ads/voiceovers/TVCs (non-narrative) → use doubao-creative-video; Pure 2D animated comic final cut → use ai-drama-video-production]
• 🔗 Upstream/Downstream / Recommended Pairings: Upstream [blockbuster-script-director script adaptation / ai-story-vault topic & hit research / video-generation-preferences user constraints] → This Skill (Live-Action Short Drama Full Pipeline Controller) → Downstream [script-asset-master art assets / image-reference reference image standards / jimeng-studio·jimeng-director Jimeng engine deployment / jianying-editor post-production editing]
• 🧭 Boundaries with Adjacent Skills: This skill is a "live-action short drama full pipeline controller," spanning script → storyboard → prompt → generation → review → hard-cut assembly → delivery; it has 33 references including No.23 cross-task routing (26–29 are storyboarding operation/recipe/coherence/AI deployment four-piece set, 30–31 are action scenes/confrontation·style-lock·long-scene·timestamp rendering two special topics, 32 is user post-production sound/music reference, 33 is group-end technical panorama/5-second technical patch chain anchor point); it differs from jimeng-studio, which is more "Jimeng Engine Short Drama Studio Execution," while this suite is a more complete full-process orchestration and control; pure dynamic comic/2D final cuts should use ai-drama-video-production, not the main path of this suite.
• 📦 Standard Delivery Responsibilities (Regular Two Items + MJ On-Demand): As the short drama controller, this skill routinely leads ① Project Outline + ② General Model Asset Prompts (② specific art assets can be led by script-asset-master, with this suite orchestrating deployment); ③ MJ (Midjourney) Prompts are on-demand items – generated only when you explicitly mention/specify using Midjourney (saying "MJ" or "Midjourney"); I do not proactively output, include by default, or pre-write them. When you mention MJ, I will fully output English prompts per MJ standards (--ar/--niji/--stylize/--weird/--chaos/--seed/--cref/--sref, including rationale for 3D values). Template paths and three control dimension specifications are in the "📦 Standard Delivery" section below.
• ⚠️ Fixed User Specifications: Final cut has no subtitles, no BGM, TTS is disabled (dialogue is generated directly by the video model); before each video generation, first fully display the detailed prompt, and generate the video simultaneously with the prompt display (do not stop to wait for confirmation); deliver the final cut together with the full prompt for review; character images must be referenced with @, enforce image reference, do not generate characters with pure text; user-provided character images: one person per role, no duplicates; extra roles are pure text descriptions; after generating each video, you must present the "final cut + that video's complete script/storyboard + complete video prompt (@ reviewed assets)" to the user via an embedded preview player; generate the next only after user approval, generate and review one by one; unauthorized continuous batch automatic output of all clips is prohibited (modules still advance step-by-step with confirmation, no default full automatic run); also attach a public access link; art assets can be directly generated by AI, but each batch of images must be reviewed before finalization and @ reference; user self-made is optional; dialogue must be written into prompts, do not use TTS; no subtitles throughout the entire final cut; only hard cuts/hard assemblies throughout the entire film, do not use any transition effects; no terminal usage; the entire workflow is completed in the chat interface; the entire skill does not contain any watermark removal related content; final cut only performs lossless hard assembly, no active editing (post-production editing is handled by the user); absolutely zero subtitles throughout the entire film, do not write any on-screen text instructions in prompts.

🔀 This Skill's Live-Action / Comic Drama Branches

Branch	Live-Action or Comic Drama	Trigger / Style Engine	Corresponding Prompt Style
Default Main Branch · Realistic Live-Action Short Drama	Live-Action Realistic Oriented	Default model_version=seedance_2.5 (realistic highest priority); Jimeng "Horizontal Screen Film Texture (ARRI ALEXA 65)"; No.21 Fuji Film Documentary Look Preset	Live-Action Realistic: realistic human proportions, realistic skin and lighting, camera texture; write prompts with "documentary photography texture/cinematic feel/film grain"; MJ uses --v 6.1, do not add --niji
Stylized / Dynamic Comic Branch	Comic Drama Dynamic Comic Oriented	Enabled when user explicitly requests "anime/dynamic comic/illustration/Chinese ink painting/comic style"; Switch to Kling stylization engine, dynamic comic engine, or MJ --niji 6	Visual style segment changes to "dynamic comic/illustration/cel shading/hand-drawn background"; MJ adds --niji 6 --stylize 200; ② asset card style segment is switched synchronously and noted

Note: The main category of this suite is judged as Live-Action Realistic Oriented based on "default/primary output" (Seedance 2.5 default realistic, horizontal screen film texture); the second row above is the on-demand stylized/comic drama branch – when enabled, you must: ① check "Comic Drama Dynamic Comic Oriented" in Section 0 of the Project Outline, and ② switch the style segment in the asset prompts (if you have already mentioned MJ, then ③ switch synchronously); do not mix with realistic vocabulary.

⛔ Phase Gate Control General Principles (Option-Driven · User Confirmation · Mandatory Review of AI-Generated Assets + Mandatory Viewing of Video Final Cuts)

This suite no longer automatically runs end-to-end full short dramas by default. The standard paradigm is fully documented in /home/user/Doubao/chats/38436370915409154/_工作规范/阶段门控总则（标准范式）.md; the following are the implementation points for this suite, taking precedence over any older descriptions like "automatic continuous / one-stop / automatic asset generation" below.

0. Supreme Principles (Cannot Be Violated)

1. Prohibition of One-Time Automatic Complete Production: When a user provides only a single idea, the system must not automatically complete script → assets → storyboard → video → post-production all at once.
2. Three Steps Must Be Completed at the End of Each Phase: ⛔ Pause → Give Options (each with a one-sentence highlight + pros & cons + ⭐ recommended item) → Wait for explicit user agreement; only then proceed to the next phase; never proceed without user confirmation.
3. At each gating point, the user can: Reply with a number to select / mix and match multiple options / reply "more ideas" for a new batch / directly write modification suggestions.
4. 【Art Assets: AI Can Directly Generate, but Each Image Must Be Reviewed Before Finalization @】 : Character, scene, prop, key frame / storyboard reference images that enter the final cut can be directly generated by AI using text-to-image (T2I) ; however, after each image / batch is generated, you must immediately present the "completed image + the general model prompt used for that image + 3D parameters" to the user for review (if you mention MJ, also attach the MJ version English prompt) – the user can ✅ directly approve / provide modification suggestions / request regeneration – only after approval is it finalized, and then it can be referenced with @ in storyboards and video prompts, with mandatory image reference; if not approved, modify the prompt / regenerate based on feedback, review again, and finalize only upon approval. "User self-made" remains an optional path: when the user wants to create their own images, the AI will provide general model prompts + 3D + list order + suggestions by default (additionally provide MJ version if you mention MJ); upon the user's return, it still goes through the same review confirmation process.
   • AI remains responsible for video generation (T2V / I2V) : Before each video generation, first fully display the detailed prompt for that video in the conversation, and directly call the generation simultaneously with the prompt display (do not stop to wait for confirmation, do not set a confirmation step) ; after the final cut is generated, present the "final cut (embedded preview player) + the complete script/storyboard for that video + the complete video prompt (@ reviewed assets)" all together for the user to see; proceed to generate the next only after user approval.
5. Any asset image output must be delivered for user review; do not finalize on your own: AI can directly generate asset images, but as soon as an image is generated, it must be immediately delivered for review; do not batch them, do not unilaterally finalize; reference / schematic images intended to be used as finalized assets in the final cut must also pass user review.

1. Unified Output Format for Each Gating Point (Provide Exactly This, Do Not Omit)

### ✅ This Phase Complete
(1–3 sentences explaining what this step produced)

### Please Choose (N Directions Total)
| Number | Plan | One-Sentence Highlight | Pros | Cons |
|---|---|---|---|---|
| ① | 〔　〕 | 〔　〕 | 〔　〕 | 〔　〕 |

⭐ **My Recommendation: Item ①** – Reason: 〔　〕

### ⛔ Paused; Please Reply Before I Continue
- Reply "choose ①" = adopt; "①+② mix" = combine; "more ideas" = new batch; or directly write modification suggestions.
👉 I will proceed to the next phase only upon receiving your explicit selection / confirmation.

2. Standard Gate Control Rhythm (Common for Live-Action / Comic Drama, Wait for Confirmation at Each Gate)

Gate	Phase	Options / Actions for User	Passing Condition
Gate 1	Story Ideation	When the user only has an idea, first present 3–5 differentiated story directions (varying in genre / core setting / protagonist archetype / main conflict / opening hook), filled into the Story Ideation Options Table	User selects / mixes / requests more and confirms the story
Step 0 (Asset Phase)	Read script-asset-master (P0)	Before entering any asset phase, first fully Read the SKILL.md of script-asset-master; do not skip, do not execute based on memory	Completed reading before entering asset plan
Gate 2	Character / Asset Plan Selection	Provide plan options (only define the plan) for character count & profiles, scene configuration, prop list, key frame direction, etc., filled into Section A of the Asset Review Confirmation Table	User selects / modifies and confirms the asset plan
Gate 3	Mandatory Review for AI-Generated Assets (Critical, Review Each of Four Categories by Gate)	AI can directly call T2I to generate characters / scenes / props / key frames, simultaneously producing general model version prompts + 3D parameters + asset list & production order + production suggestions (additionally provide MJ version if you mention MJ); after each image/batch, immediately present the "completed image + prompt + 3D" to the user for review	User approves (✅/provides comments/requests regeneration) before finalization and @ reference; user can also choose self-made (returned assets also undergo review)
Gate 4	Video Generation	AI first fully displays the video prompt, and directly generates the video simultaneously with the prompt display (do not stop to wait for confirmation) ; after the final cut, present the "final cut (embedded player) + complete script/storyboard + complete video prompt (@ reviewed assets)" together for the user to see	User approves this final cut before generating the next, generate and review one by one; do not unauthorized continuous batch output
Gate 5	Post-Production / Delivery	Options for hard-cut assembly, delivery specifications, cover image (AI can generate cover, but generated images must be reviewed), etc.	User confirms item by item

Do not enter video generation until all characters / scenes / props / key frames have been reviewed and finalized; generate videos one by one, review and approve each before outputting the next; do not batch them together and automatically run all at once.

3. Story Ideation Operation Method (Place at the Very Beginning of Module 1)

• When the user only has an idea / theme / one-sentence story, do not write an outline, do not define assets first; produce 3–5 differentiated directions at once, filled into the Story Ideation Options Table (project directory /home/user/Doubao/chats/38436370915409154/故事发散选项表.md): Number / Direction / Genre / Core Setting / Protagonist Archetype / Main Conflict / Opening Hook / One-Sentence Highlight / Pros / Cons / Recommendation.
• Each direction must be clearly different in genre, core setting, protagonist archetype, main conflict, and opening hook; after presenting, ⛔ pause and wait for confirmation.
• After the user confirms the story, only then produce / finalize ① the Project Outline.

3+. Visual Hook & Causal Chain · Two Mandatory Creative Requirements

• ① Hook as a Visual (Show, Don't "Tell"): The opening hook must be a concrete, visible image / action / scene, capturing attention through the first frame or first few seconds via visual impact or visual suspense; do not rely on voiceover, dialogue, subtitles, or titles to "tell" the hook; the hook image must directly relate to the core conflict / suspense; do not use irrelevant eye-catching gimmicks. In story ideation and the outline, the "Opening Hook" field must always be filled with a specific hook image, not an abstract concept.
• ② Causal Relationship (A→B→C, Eliminate Mundane Chronology): The plot advances via a causal chain, not a series of "and then... and then..." events; the anomaly / suspense introduced by the hook must have a cause (can be revealed later); every action by a character must have a motivation and a consequence; foreshadowing is the "cause," payoff is the "effect"; twists must be surprising yet plausible; when self-checking, ask for each scene: "What causes it, and what does it lead to?" Delete paragraphs that have no causal relationship with the main storyline.

4. Mandatory Review Operation for AI-Generated Assets (Critical, User Self-Made Optional)

⛔ P0 Must-Read: Before entering any asset phase, you must first fully Read the SKILL.md of script-asset-master (live: /home/user/.doubao/agent_mode/workspace/.user_skills/script-asset-master/SKILL.md; stable copy: /home/user/Doubao/chats/38436370915409154/已优化技能/script-asset-master/SKILL.md). Do not skip, do not execute based on memory. Fixed order for asset phase: Read script-asset-master (P0) → Present asset plan options → User confirms plan → AI can directly call T2I to generate assets (simultaneously provide general model version prompts + 3D parameters + asset list/order + production suggestions; additionally provide MJ version only when you mention MJ) → After each image/batch, immediately deliver "completed image + prompt + 3D" to the user for review → Only after user approval is it finalized and @ referenced.

• Asset Plan Selection (Only Define the Plan): Based on the confirmed story, provide options for character count & profiles, scene configuration, prop list, key frame / storyboard directions (Section A of the Asset Review Confirmation Table); user selects / modifies and confirms.
• AI's Responsibilities in the Asset Phase (Can Generate Images + Must Deliver for Review): ① Can directly call T2I to generate assets based on the confirmed plan; after generating images, immediately deliver them for user review; do not batch, do not unilaterally finalize; ② By default, deliver one set of [General Model Version Prompts] (in Chinese, compatible with Seedream / Jimeng / Kling / Seedance); [MJ Version Prompts] (in English, including --ar / --niji / --cref / --sref) are additionally produced only when you explicitly request Midjourney, not attached by default with the generated image; ③ Three control dimension parameters (stylization --stylize, weirdness --weird, diversity --chaos + fixed --seed) and rationale for values – 🎥 Live-Action Realistic = low / low / low + fixed seed; 🎴 Comic Drama Strong Style = stylization mediumhigh, weird low, diversity lowmedium + fixed seed; ④ Asset list & production order (recommended: character anchor → scene empty shot → props → key frame / storyboard reference image) and production suggestions (shot size / lighting / pose, four-grid / three-view / empty shot, lock consistency via fixed seed & --cref·--sref, etc.).
• ✅ AI can directly use image generation tools to produce assets, but after each image/batch is generated, it must immediately be delivered for user review and only finalized upon approval; script-asset-master in this context is positioned as "can directly produce assets (prompt → call generation) + deliver images for user review and confirmation"; it continues to retain the ability to provide general model version prompts, 3D, asset list/order, and production suggestions (MJ version prompts are only produced on demand when the user mentions MJ); for replication tasks, it still produces / generates a full set of "remade" assets (no copying from original, high-fidelity shot-by-shot), but each image must also be reviewed.
• After user approval (✅ / provides modification suggestions / requests regeneration), it can be referenced with @ in storyboards / video prompts, with mandatory image reference; confirm one category at a time. Record in the Asset Review Confirmation Table (project directory /home/user/Doubao/chats/38436370915409154/资产审核确认表.md), tracking per asset: "AI generated pending review / user self-made / needs rework / reviewed and can be @ referenced". "User Self-Made" is an optional path: when the user creates their own images, AI still provides the materials; upon the user's return, it also goes through the same review process.

5. Regular Two Items Produced in Batches According to Gate Rhythm (MJ On-Demand, Not Pre-Written)

• ① Project Outline: Produced / finalized only after "story confirmation".
• ② General Model Asset Prompts: Provided in batches according to the user's generation progress – character prompts given after character confirmation, scene prompts after scene confirmation, props/key frames likewise; not provided all at once in advance.
• ③ MJ (Midjourney) Prompts: Not pre-written, not attached with ②; only produced after you explicitly mention MJ / Midjourney, within the scope you request and following the same batching rhythm.

6. Relationship with Existing Standards (Overlay, No Conflict, No Deletion) Retain and overlay: 🏷️ Live-Action / Comic Drama Classification Labels; 📋 Six-Point Script Self-Check; 📦 Standard Delivery (Regular Two Items + MJ On-Demand); 🎛️ Asset 3D; ⚠️ No Subtitles, No BGM, TTS Disabled, Mandatory @ and Image Reference for Characters/Assets. Compared to the old version, this general principle changes two points: ① Phase/module advancement changed from "automatic continuous running" to "pause at each gate, give options, wait for confirmation"; ② Asset production changed from "all finalized assets are user self-made, AI only provides prompts/parameters/lists/suggestions, does not generate images" to "AI can directly generate art assets, but each image/batch must be reviewed before finalization @; user self-made is optional". All other capabilities are retained with zero deletion; video generation (T2V/I2V) is still executed by AI, and each time, first fully display the prompt, generate simultaneously with the prompt display (do not stop to wait for confirmation), then present the final cut + complete script/storyboard + complete prompt to the user; proceed to the next only after approval.

🧭 Handover Master Guide (Read This First for Any New AI Handover, Follow Linearly)

This block is the authoritative guide: what to read first, where to enter, in what order to read files, what to produce, when to stop and wait for confirmation. The sections below "Full Workflow Sequence / Mandatory Reading List / Eight Modules / Global Iron Laws / Self-Check Gates / Reference File Index" are the detailed expansion; if this block conflicts with the content below, this block and the Global Iron Laws take precedence.

① Mandatory Every Time (Do Not Start Work Without Reading): Full Read this SKILL.md – Skill Extension Specification + Global Iron Laws 0–39 + This Roadmap; For new projects, also Read references/16-project-playbook.md (Project Startup Checklist).

② Determine Entry Point (See What the User Provided, Cut into the Corresponding Module; If Requirements Are Unclear, Ask Only One Short Question, Do Not Present a Multi-Option Menu)

User Provides	Entry Point
Only an idea / theme / one-sentence story	Module 1 Script
Already have a script / story excerpt	Module 2 Storyboarding
Already have a storyboard	Module 3 Prompts
Already have prompts, need to batch generate videos	Module 4 Generate Video
Already have video clips, need review / fixes	Module 5 Review + Consistency
All shots confirmed, need to assemble into one sequence	Module 6 Hard-Cut Assembly (No Editing)
Need cover / release specs / delivery	Module 7 Delivery
Full 0-to-1 package	1→2→3→4→5→6→7
Non-short drama (commercial videos / material processing / online video extraction / audio / image assets)	23 Cross-task Master Router

③ Reading Route Map (Enter Phase → Read in Order → Produce Output → Pass Confirmation Gate → Next Phase)

Phase	Read in Order Before Starting	Phase Output	→
Module 1 Script	New project: first 16 → Select formula per ④ (09/11/20, can overlay 10) → 01	Completed script (duration determined by story, not fixed)	M2
Module 2 Storyboard	26 (Main line: Scene Card→Beat→Coverage) → 27 (Apply Recipe) → 28 (Coherence) → 29 (AI Deployment); Add 30 for action scenes, add 31 for confrontation/style lock/long scene/timestamp; 15 (if characters>2 or scenes>2), 25 (top view), 02 (red lines), 14 (terminology); Output container: apply 22; Multi-shot same scene: use 24 Camera Base Map method; Same scene across groups, multi-person continuous movement/action handover, need to lock "group-end dynamic state": additionally read No.33, and first select Mode A (group tail 1 second) or Mode B (5-second technical patch)	Storyboard script / storyboard groups (flexible 5–30 seconds)	M3
2.5 Assets (Critical Gate, Mandatory Review of Generated Images)	If script-asset-master is available, it produces the asset prompt package + 3D + list order + suggestions, and can directly call T2I to generate characters/scenes/props/key frames; after each image/batch, immediately deliver for user review; finalize @ only after approval (user self-made optional, returned assets also reviewed)	User-reviewed and finalized reference images, confirmed item by item	M3
Module 3 Prompts	03 → 14 (For ready-made documentary style, directly apply 21); Inherit 22 storyboard groups	Prompts containing @ images + reference images	M4
Module 4 Generate Video	04 (For multi-shot same scene, first lock camera base map using 24; for cross-group continuation, obtain technical panorama anchor frame using 33; if Mode A cannot precisely trim the tail, switch to Mode B)	First fully display the prompt, generate simultaneously with display (do not stop to wait for confirmation): final cut (embedded preview) + complete script/storyboard + complete video prompt (@ reviewed assets) all presented to user → shot-by-shot self-check (unexpected text/morphing/face/clothing/positioning/axis/audio-video sync) → proceed to generate next only after user approval	M5
Module 5 Review	05 (＋15)	Review + consistency report, repaired material	M6
Module 6 Hard-Cut Assembly	06	Lossless hard-cut rough final cut (raw, unprocessed, delivered for user self-editing)	Delivery
Module 7 Delivery	07 → 13 (Cover)	Original quality file + public link + multi-specs + cover	—

④ How to Choose the Default Methodology (Select one main structural formula based on genre; 10 Emotion / 14 Terminology / 21 Visual Style / 22 Template are enhancement layers, can be stacked, no conflict)

Genre Characteristics	Default Application
Urban Fantasy / Romance / Supernatural / Superpowers	09 Structure + 10 Emotion
Period/Historical / True Events / Major Diplomatic Events / Big Figure Small Story	11 (In practice with 12, can overlay 10)
Oppression Rescue / Liberation / Serf Old Society / Rescuers / Strong Conflict	20 (can overlay 10; visual default 21)
Need ready-made documentary / period / film grain look (visual texture)	21 (any genre can overlay; terminology principles check 14)
Storyboards (any genre, Module 2 operation)	26 Main Line → 27 Apply Recipe → 28 Coherence → 29 AI Deployment (red lines check 02, container apply 22)
Positive Action Scenes (fights/chases/gunfights)	30 (Causal grammar/action unit vs. shot/multi-person; AI boundary still check 29)
Confrontation Dialogue Scenes / Lock Full-Film Style / Long Scenes / Shot-by-Shot Second-Level Timestamps	31 (Style Six Slots/Asset Anchor Points/Division Continuation/Rendering; Container apply 22)
Standard Storyboard Group Output (Module 2→3 Universal Format)	22

⑤ Confirmation Gates (Two Rhythms, Don't Mix Them Up)

• Between Modules: Upon completing each module, first perform a complete self-check and display; proceed to the next module only when the user says "ok"; if the user provides modifications, revise and perform the self-check again.
• Within Module 4 Video Generation: For each video, first fully display the prompt, generate simultaneously with the prompt display (do not stop to wait for confirmation); after the final cut is ready, present the "final cut (embedded preview) + complete script/storyboard + complete video prompt (@ reviewed assets)" all together to the user → perform a shot-by-shot self-check and display (unexpected text/morphing/face/clothing/positioning/180° axis/audio-video sync) → wait for user approval before generating the next; do not default to automatically running all clips continuously in batch (the user can interrupt at any time with modifications; if inappropriate, revise per feedback/regenerate and review again).
• Module 6 Only Does Hard-Cut Assembly, No Editing: Post-production editing / music / color grading are handled by the user; real ambient sounds (footsteps, rain, wind, etc.) are retained as-is; only artificial BGM is prohibited (Iron Law 26); this skill delivers a lossless rough final cut that can be directly imported into editing software, not packaged as a "completed final cut".

⑥ Synchronize Documents After Any Changes: Any new additions/modifications or version updates must be synchronously reflected in the README's "Changelog" (version, date, content); documentation is considered complete only when consistent with the skill before submission (see Skill Extension Specification "Four Things" below).

---

📐 Skill Extension Specification (Follow This for Any Future Additions)

Directory Structure:

ai-drama-master-suite/
├── SKILL.md              ← Master scheduler, contains only rules and index, no specific content
└── references/
    ├── 01-07             ← Process specifications (in module order)
    ├── 08                ← Case study references
    ├── 09-10             ← Hit methodology
    └── 11+               ← Future additions continue numbering

Four Things That Must Be Done When Adding New Files:

1. Create a new independent .md file, continuing the numbering sequence (11, 12, 13…), naming format number-short English/pinyin.md
2. Add one line in the corresponding group of the "Reference File Index" at the bottom of SKILL.md, clearly stating: what the file is and when to read it
3. If the new file is a methodology to be used when writing scripts, add trigger words and loading instructions in the "Must Read Before Script Writing" block at the top of this file
4. Synchronously update README: Every time content is added/modified or there is a version update, the README's "Changelog" (version number, date, update content) must be synchronously updated; the README and skill content must be consistent before it is considered complete and can be submitted

Things Not To Do:

• ❌ Do not append large sections of content to existing files (polluting the original structure)
• ❌ Do not write specific creative methodology in SKILL.md (SKILL.md only contains rules and index)
• ❌ Do not change file names or numbers (keep references stable)
• ❌ Do not duplicate content between new and old files

Grouping Rules:

• Group A (01-07): Module process specifications, in order of modules 1-7
• Group B (08): Specific case study references (how a specific work was made)
• Group C (09+): Reusable creative methodology (default rules to apply when writing scripts in the future)

---

🚨 Full Workflow Sequence (Strictly Follow This, Do Not Skip Steps)

Each time you receive a task, first follow this sequence: First determine where to start, then read files in order and work.

Step 0: Read this SKILL.md (Global Iron Laws 0-39, mandatory)
  │
  ├─ User only has an idea / wants to write a script ──→ Module 1
  │     Read: 09 hit formula + 10 emotion design + 01 script standard
  │
  ├─ Already have a script, need storyboard ──→ Module 2
  │     Read: 02 storyboard standard + 14 visual terminology
  │
  ├─ Storyboard complete, need reference images ──→ Module 2.5 Asset Preparation (Critical Gate Step)
  │     ★ Art assets: AI can directly generate: AI can directly call T2I, simultaneously produce general model version prompts + 3D parameters + asset list & production order + production suggestions (MJ version only when you mention)
  │     ★ If *script-asset-master* (Full Asset Master) is installed, it produces the above prompt package and can directly generate images; scene empty shots/character four-grid/props/key frames can be directly generated by AI
  │     ★ After each image/batch, must immediately deliver the "completed image + prompt + 3D" for user review; finalize `@` only after approval; if user wants self-made, they generate images, return them, and they undergo the same review
  │     ★ Only after user review and finalization can they be `@` referenced in prompts; do not enter Module 4 Video Generation until all assets are reviewed and finalized
  │
  ├─ Have storyboard and reference images, need to write prompts ──→ Module 3
  │     Read: 03 prompt standard + 14 visual terminology
  │     Prompts must `@` reference images for characters
  │
  ├─ Have prompts, need to generate video ──→ Module 4
  │     Read: 04 video generation standard
  │     First fully display prompt, generate simultaneously with prompt display (do not stop to wait for confirmation), deliver final cut together with complete script/storyboard + complete prompt to user and perform shot-by-shot check
  │
  ├─ Have video, need to review ──→ Module 5
  │     Read: 05 quality review
  │
  ├─ Shots all confirmed, need to assemble into one sequence ──→ Module 6 Hard-Cut Assembly and Delivery (No Editing)
  │     Read: 06 hard-cut assembly standard (post-production editing handled by user)
  │
  └─ Need to deliver / create cover ──→ Module 7
         Read: 07 delivery standard + 13 cover design standard

Quick Reference: Mandatory Reading By Phase

Phase	Mandatory Files	Consequences of Skipping
Entire Process	This SKILL.md (Iron Laws 0-39)	Violate global rules
Writing Script	Select by genre 09/11/20 (can overlay 10) + 01; To develop outline into full script, additionally read 35 (M1 Screenwriting/M2 Novelized dual writing modes + tempo gear four-speed); For script doctor on outline/episode/scene, read 36; For line-by-line dialogue polish, read 37; For opening hook and full-film tempo curve, read 43 (3-second hot open/10-second background setup/six-beat rhythm, expanded in Iron Law 41)	Script is flat, lacks emotional impact; scene writing method and tempo gear undefined; dialogue is generic, characters don't sound authentic; hook fails to retain audience, emotion is monotone, middle section collapses
Storyboarding	26 Main Line + 27 Recipe + 28 Coherence + 29 AI Adaptation + 30 Action Scenes + 31 Confrontation/Style Lock/Long Scene/Timestamp; 02 Red Lines + 14 Terminology (Container apply 22; Multi-shot same scene additionally read 24; Spatial planning read 25; Cross-group lock group-end dynamic state read 33; For specific action genre, read 38; For cross-group text script notes/first shot matching, read 39); For selecting camera movement/shot size/camera placement for each scene, check 42 Camera Language Toolbox (practical version of No.14 terminology)	Shots lack system, cannot break down scenes, output breaks; action genre has no specialized technique; group-end states don't connect; don't know whether to push or pull, tilt or look down for a given scene
Asset Preparation (Critical Gate)	If script-asset-master is available, read its SKILL.md (can directly produce assets + prompt package, generated images must be reviewed); AI can directly generate asset images, generate and immediately deliver for user review and finalization (user self-made optional); For extras direction/prop orientation & single-instance lock, read 40 (protagonist main scenes still based on script-asset-master + 15)	Images generated without review are unilaterally finalized and @ referenced; user-unauthorized assets enter video; extras turn their backs or block the camera, props inexplicably duplicate
Writing Prompts	03 + 14 (Documentary style apply 21; Inherit 22; For pure text body writing / ultra-long deduplication & splitting, read 41)	Prompts are unprofessional; management jargon mixed into body text, ultra-long text forced into one group
Generating Video	04 (For multi-shot same scene, additionally read 24; For cross-group continuation, read 33 and confirm Mode A/B)	Wrong process
Reviewing	05 (For multi-shot same scene, additionally read 24; For cross-group consistency, use No.33 anchor frame for verification)	Miss issues
Hard-Cut Assembly Delivery	06 (Mode A: remove each group's last 1-second anchor tail before splicing; Mode B: 5-second technical patch segments not in filelist)	Re-encoding reduces quality / audio-video desync / unauthorized editing
Cover Design & Delivery	07 + 13	Cover looks like a screenshot with text

Execution Iron Laws

1. Asset Preparation is a Critical Gate Step (AI-generated assets must be reviewed, user self-made optional): AI can directly call T2I to generate character, scene, prop, and key frame art assets; after each image/batch is generated, immediately deliver the "completed image + used general model prompt + 3D" for user review (additionally provide MJ version when you mention MJ); finalize and @ reference only after approval, with mandatory image reference; if not approved, modify the prompt/regenerate and review again. AI (including script-asset-master) by default produces general model version prompts + 3D parameters + asset list & production order + production suggestions; ③ MJ prompts are only produced on demand when you explicitly mention them. Do not enter video generation until all assets are reviewed and finalized.
2. Before entering any module, fully Read the corresponding file with the Read tool before working; do not guess based on the file name.
3. Before writing prompts, you must first browse the No.14 Visual Terminology Table; do not just write the word "cinematic".
4. Before creating a cover, you must first browse the No.13 Cover Standard; do not just add text to a screenshot.
5. If a file cannot be read/does not exist, immediately inform the user; do not fabricate.

---

🚨 Mandatory Reading File List (By Phase, Must Be Fully Read, No Jumping/Impression/Fabrication)

During the corresponding phase, the following files must be fully read using the Read tool before starting work. Not having read means not having read; do not guess the content based on the file name, do not read only half, do not start working with a "general idea." Each file has specific rules and parameters, and skipping reading will directly make the output unqualified.

P0 Level – Must Check Every Time (Highest Weight, No Work Without Reading)

File	When Must Read	Why Not To Skip
This SKILL.md	Every time you enter this skill	All Global Iron Laws 0-39 are here
16-project-playbook.md	Every time a new project starts	Project startup checklist + format transition guide; skipping causes character model drift/scene jumps
14-cinematic-vocabulary.md	Every time you write video prompts or storyboards	Shot size/focal length/lighting/camera movement/color terminology; skipping makes prompts unprofessional
13-cover-design.md	Every time you create a cover	Movie poster style cover standard; skipping turns it back to a screenshot with text

P1 Level – Must Read When Entering the Corresponding Module

File	When to Read	Consequences of Skipping
09-hkkk-formula.md	Before writing script in Module 1 (fantasy/supernatural/superpower genre)	No hit structure, script is flat
34-viral-family-drama-formula.md	Before writing script in Module 1 (reality family/family ethics/social pain point genre, default track)	Hook fails to retain, story lacks impact, no discussion point
10-emotion-design.md	Before writing script in Module 1	Emotion lacks impact, audience won't cry
01-script-creation.md	When writing script in Module 1	Script format and specifications are wrong
35-script-drafting-modes.md	When developing an outline into a full script in Module 1 (M1 Standard Screenwriting/M2 Novelized Narrative dual modes + orthogonal episode split switch + four-speed tempo gear)	Scene writing method undefined, tempo gear selected incorrectly, overall drama rhythm fluctuates erratically
36-script-doctor.md	When needing a script doctor for outline/episode/scene/complete script in Module 1 (user says "diagnose/look at what's wrong with the script/fix the script")	Script problems get glossed over, cannot grasp must-fix points, fabricate issues just to fill the form
37-dialogue-expert.md	When polishing dialogue in Module 1 (user says "change this line/dialogue is boring/subtext/tone is wrong")	Dialogue is generic, does not sound like authentic human speech
38-action-scene-guidelines.md	When the genre involves action (fights/shootouts/chases) in Module 1 or 2	Action scenes lack specialized techniques, cannot power up the length
02-red-lines.md	When writing storyboards in Module 2	Breaking 180° rule, continuity errors, jump cuts
14-cinematic-vocabulary.md	Same as P0	Prompt vocabulary is poor
26-storyboard-master.md	When writing storyboards in Module 2 (main line)	No systematic storyboarding method, don't know which shot to put where
27-storyboard-recipes.md	When applying a storyboard formula in Module 2	Storyboard rhythm is off, shot variety is poor
28-coherence.md	When checking storyboard coherence in Module 2	Characters' positions jump, screen direction confused, props lose consistency
29-ai-storyboard-deployment.md	When converting storyboards to video prompts in Module 2	Storyboards cannot be executed by AI, generated shots completely different from intended
40-extras-direction-and-prop-lock.md	When planning extras direction or prop direction/single-instance in Module 2	Extras randomly turn their backs to the camera/cluster in the middle of the frame; props inexplicably duplicate or disappear between shots
15-model-card-standard.md	When the story involves >2 characters or >2 scenes in Module 2	Characters look the same, scene style is inconsistent
24-multi-shot-same-scene-camera-base-map.md	When a scene has multiple shots in Module 2–4	Camera axis jumps wildly within the same scene
30-action-scene-causal-grammar.md	When there is a positive action scene (fights/chases/gunfights) in Module 2	Action shots lack power and logic, characters randomly teleport
31-confrontation-style-lock-long-scene.md	When it's a dialogue confrontation scene / needs to lock a single style throughout the film / is a long continuous scene with >15 shots in Module 2	Confrontation shots are flat; style jumps/switches randomly between groups; you don't know how to split a long scene into groups
33-group-end-technical-panorama.md	When cross-group continuation is needed (e.g., scenes A→B→A, or the same scene has multiple groups with a time gap) in Module 2–4	Cross-group state breaks, characters' positions/props/environment don't match
39-cross-group-script-notes.md	When cross-group scenes need text script notes/first shot reconciliation in Module 2	Script notes are not detailed enough, first shot matching is blind, causing group-end position breaks
41-pure-text-prompt-and-ultra-long-dedup.md	When writing prompts in Module 3	Management jargon mixed into body text; when text is long, deduplication & splitting method is wrong, causing prompt truncation or inconsistencies
03-prompt-writing.md	Same as above	Prompt format is wrong
21-film-vibe-grid.md	When needing a specific visual style in Module 3	Visual style is undefined, prompts lack texture
22-storyboard-group-container.md	When outputting storyboard groups in Module 2–3	Output format is chaotic, not standardized
04-video-generation.md	Before generating video in Module 4	Wrong generation process, missing mandatory steps
42-camera-language-toolbox.md	When choosing camera movement/shot size/camera placement for each scene in Module 2 or 3	Don't know whether this scene should push or pull, tilt or look down
05-quality-review.md	When reviewing generated video in Module 5	Missing review items, letting issues pass
06-hard-cut-assembly.md	When splicing in Module 6	Re-encodes and degrades quality, or unauthorized editing
07-delivery-standard.md	When delivering in Module 7	Wrong delivery format
43-hook-and-tempo-curve.md	When writing script/storyboard for Module 1 or 2	Hook idea is not executable; full-film tempo has no plan, rhythm is messy
13-cover-design.md	Same as P0	Cover downgrades from poster grade to screenshot

P2 Level – Must Read When the Specific Topic is Involved (Can Be Learned Anytime)

File	When to Read	Consequences of Skipping
08-production-of-ref-video.md	When user wants to replicate a reference video in Module 1–3	Shot-by-shot breakdown logic is wrong, replication fidelity is low
25-production-design-blueprint.md	When storyboarding or writing prompts for scenes with complex spatial layout in Module 2/3	Characters walk through walls or pass through walls in the edited final cut
11-real-history-event-formula.md	When script involves real historical events/genuine historical figures in Module 1	Event core, character archetypes, and era terminology are wrong, causing historical authenticity issues
12-social-topic-formula.md	When script involves social hot topics/pain points in Module 1	Topic point is flat, cannot generate discussion
20-hardship-to-salvation-formula.md	When script involves oppression rescue/liberation/tribulation/salvation in Module 1	Emotional impact falls short, climax is not strong enough
17-voiceover-music-reference.md	When user needs post-production sound effects/music in Module 5+	Sound design direction is wrong, does not match the film's style
23-cross-task-master-router.md	When the user input clearly does not fit the short drama vertical domain (e.g., commercial video, material processing, online video extraction, audio, image assets)	Processing non-short-drama tasks incorrectly, messing up the native suite's file structure or workflow

---

📖 Eight Core Modules

Module 1: Script Creation

Purpose: Create a complete short drama script based on user-provided ideas or story concepts.

Input: User's story concept / hot topic / reference work. Output: Full script (can be 90 seconds or 5 minutes, supports multiple structures and genres).

Scenarios:

• Any genre (fantasy, romance, historical, family ethics, etc.)
• Five major tracks: sentiment/conflict / funny/comedy / fantasy/setup / realistic/strong / crossover/niche
• Nine structural prototypes (e.g., 09 Hit Formula, 11 Real History Event Formula, 20 Hardship to Salvation)
• Four-act timing system: Opening Introduction (hook) → Conflict Development → Climax Turning Point → Ending Payoff

Workflow:

1. Confirm story direction: If user only has a rough idea, first use the Story Ideation Options Table to provide 3–5 differentiated story directions; confirm direction after user selects.
2. Determine structure: Select a structural formula based on genre characteristics (default: 09 fantasy/romance/supernatural; 11 historical/real events; 20 oppression rescue).
3. Write complete script: Follow the Project Outline template to produce Scene | Beat | Dialogue | Visual Description | Duration.
4. Self-check: Six-point script self-check (Hook → Setup → Conflict → Climax → Payoff → Ending).
5. Confirmation gate: ⛔ Pause → Present options → Wait for user confirmation before proceeding.

Module 2: Storyboard Design

Purpose: Break a script or narrative into visual shot sequences.

Input: Script / story excerpt / reference video. Output: Professional storyboard group (shot-by-shot breakdown, formatted).

Characteristics:

• Professional cinematic language constraints (shot size/focal length/camera movement/lighting)
• Standard storyboard group format (Group number/Shot number/Shot size/Action/Duration/Transition)
• Three-piece set: 26 Storyboard Master (main line: scene card → Beat → coverage) → 27 Apply Recipe → 28 Coherence Check → 29 AI Deployment Adaptation
  • For action scenes: add 30 Causal Grammar
  • For confrontation/style lock/long scene/timestamp: add 31
• Multi-shot same scene: use 24 Camera Base Map method to lock axis consistency
• Cross-group continuation: use 33 Group-End Technical Panorama (select Mode A/B)

Confirmation gate: ⛔ Pause → Present storyboard → Wait for user confirmation before proceeding.

Module 2.5: Asset Preparation (Critical Gate)

Purpose: Prepare character, scene, prop, and key frame reference images as visual anchors for video prompting.

Input: Confirmed story / storyboard. Output: User-reviewed and finalized reference images (character four-grid / scene empty shots / props / key frames), recorded in the Asset Review Confirmation Table.

Workflow:

1. P0 Must-Read: Read script-asset-master SKILL.md first.
2. Asset plan selection: Provide options for character count & profiles, scene configuration, prop list, key frame directions.
3. AI generates images (mandatory review): AI can directly call T2I to generate assets based on the confirmed plan, and simultaneously produce general model version prompts + 3D parameters + asset list/order + production suggestions (additionally provide MJ version only when you mention MJ). After each image/batch, immediately deliver the "completed image + prompt + 3D" for user review.
4. User review & finalization: User approves (✅/provides comments/requests regeneration) → finalize @ reference; if user self-makes, returned assets also undergo review.
5. Record asset status: Track per asset in the Asset Review Confirmation Table.
6. Gate: Do not enter Module 4 until all assets are reviewed and finalized.

Module 3: Prompt Generation

Purpose: Generate high-quality video generation prompts based on storyboards and asset references.

Input: Storyboard groups + finalized reference images (@ reviewed assets). Output: Structured prompts containing @ images + reference images.

Features:

• Supports multiple engines:
  • MiniMax H3: Three workflows (dialogue scenes / action scenes / nine-grid composition)
  • Jimeng: Multi-style engine (film texture / documentary / anime, etc.)
  • Optional MJ (Midjourney) on-demand: English prompts with --ar/--niji/--cref/--sref
• Enforce image reference and @ character: Prompts must reference approved asset images with @; do not use pure text character generation.
• Dialogue written into prompt directly: Do not use TTS.
• No subtitles: Do not write any on-screen text instructions in prompts.

Confirmation gate: ⛔ Pause → Display complete prompts → Wait for user confirmation before generating video.

Module 4: Video Generation Execution

Purpose: Execute video generation based on prompts, manage progress and version control.

Input: Confirmed prompts + @ reference images. Output: Individual video clips (each accompanied by full prompt + script/storyboard for user review).

Execution rules (highest priority):

• Engine priority: model_version=seedance_2.5 is the highest priority default engine (for live-action realistic writing).
• Prompt verbatim transmission: Original prompt is passed through without rewriting.
• Full prompt display before generation + simultaneous generation: Before each video generation, first fully display the detailed prompt in the conversation, and directly call generation simultaneously with the prompt display (do not stop to wait for confirmation, no confirmation step).
• Enforce image-to-video generation: Must use image-to-video, @ the character, do not use pure text generation.
• Delivery after generation: After the final cut is generated, present the "final cut (embedded preview player) + that video's complete script/storyboard + complete video prompt (@ reviewed assets)" all together for the user to see.
• Generate one by one, review one by one: Only after user approval for this clip do you generate the next; unauthorized continuous batch automatic output of all clips is prohibited.
• Concurrent public access link: Attach a shareable public access link.
• Self-check: After each generation, perform a shot-by-shot self-check (unexpected text/morphing/face/clothing/positioning/180° axis/audio-video sync).
• No terminal usage: Entire workflow completed in the chat interface.

Self-check items:

• ✅ No unexpected text
• ✅ No character morphing or severe deformation
• ✅ Face matches character model card
• ✅ Clothing/accessories match prop list
• ✅ Position/screen direction matches storyboard
• ✅ 180° axis not violated
• ✅ Audio-video sync normal

Module 5: Quality Review

Purpose: Multi-dimensional review of generated videos for quality and consistency issues.

Input: Generated video clips. Output: Review report + issue diagnosis + repair suggestions.

Review dimensions (seven dimensions):

1. Visual quality: Resolution, clarity, artifacts, lighting realism
2. Character consistency: Face, clothing, body proportions
3. Scene consistency: Background, props, lighting coherence
4. Action logic: Movement continuity, action-reaction correctness
5. Dialogue sync: Lip-sync, emotional match
6. Narrative coherence: Shot-to-shot logical connection
7. Technical defects: Morphing, flickering, clipping, unexpected objects

AI issue diagnosis database: Automatically match known issue types and suggest fixes.

Self-check flow:

• Compare storyboard constraints vs actual output
• Record issue type (critical/major/minor)
• Provide repair prompts (re-generation specific segments)

Module 6: Lossless Hard-Cut Assembly

Purpose: Splice confirmed shots into a complete playable film in sequence, losslessly.

Input: User-approved individual video clips. Output: Lossless hard-cut rough final cut (no re-encoding).

Requirements:

• Hard cut/hard assembly only: Use only hard cuts throughout, no transitions, no effects.
• No subtitles: No subtitles in the entire film (editing handled by user).
• Lossless stream copy: Shots are concatenated via stream copy, no re-encoding, preserving original quality.
• Audio-video sync verification: Verify sync at each splice point.
• Original quality, zero compression: Deliver at original bitrate/resolution.
• Do not actively edit: Only perform lossless hard assembly, no editing (post-production editing handled by user).
• Mode A/B: Mode A: remove each group's last 1-second anchor tail before splicing; Mode B: use 5-second technical patch for group boundary transition, but the patch itself does not go into filelist.

Module 7: Final Delivery

Purpose: Generate multi-platform delivery packages and public access links.

Input: Lossless hard-assembled rough final cut. Output: Multi-spec files + cover image + public access link.

Specs:

• Multi-platform support: Horizontal (1920×1080) / vertical (1080×1920) / square (1080×1080).
• Cover: Use No.13 movie poster style standard, AI can generate, generated images must be reviewed.
• File format: MP4 (H.264/H.265).
• Public access link: Provide a shareable public access link.
• User self-editing note: Clearly indicate that the delivered piece is a rough cut for user self-editing.

---

🌐 Reference File Index

Group A: Module Process Specifications (Read in Module Order)

No.	File Name	Brief Description	When to Read
01	01-script-creation.md	Script writing specification, formatting, duration control	Module 1 Script
02	02-red-lines.md	Visual red lines (180° rule, match cuts, continuity)	Module 2 Storyboarding
03	03-prompt-writing.md	Video prompt writing specification, structure, terminology	Module 3 Prompts
04	04-video-generation.md	Video generation execution flow, parameter settings, engine selection	Module 4 Video Generation
05	05-quality-review.md	Seven-dimension quality review standard, issue diagnosis database	Module 5 Review
06	06-hard-cut-assembly.md	Lossless hard-cut assembly specification, stream copy parameters	Module 6 Assembly
07	07-delivery-standard.md	Multi-platform delivery format, public link generation	Module 7 Delivery

Group B: Case Studies & Reference

No.	File Name	Brief Description	When to Read
08	08-production-of-ref-video.md	Reference video shot-by-shot breakdown, replication methodology	When user wants to replicate a reference work

Group C: Reusable Creative Methodology

No.	File Name	Brief Description	When to Read
09	09-hkkk-formula.md	Hit formula: fantasy/romance/supernatural genre structure	Module 1 Script
10	10-emotion-design.md	Emotional impact point design, audience catharsis triggering	Module 1 Script
11	11-real-history-event-formula.md	Real historical event genre structure, era terminology	Module 1 Script (historical topics)
12	12-social-topic-formula.md	Social hot topic genre structure, discussion guidance	Module 1 Script (social topics)
13	13-cover-design.md	Cover design standard, movie poster style specification	Module 7 Delivery / Cover
14	14-cinematic-vocabulary.md	Visual terminology: shot size, focal length, camera movement, lighting, color	Modules 2–3 Prompts & Storyboarding
15	15-model-card-standard.md	Character/scene/prop model card standard, consistency locking	Modules 2–5
16	16-project-playbook.md	Project startup checklist, format transition guide, system prompt example	New Project Start
17	17-voiceover-music-reference.md	Post-production sound/music style reference specification	Module 5+ (when user needs)
20	20-hardship-to-salvation-formula.md	Hardship to salvation genre structure, oppression rescue theme	Module 1 Script (oppression rescue topics)
21	21-film-vibe-grid.md	Ready-made visual style library: documentary/period/film grain presets	Module 3 Prompts
22	22-storyboard-group-container.md	Standard storyboard group output format, universal for Modules 2→3	Module 2 Storyboard Output / Module 3 Input
23	23-cross-task-master-router.md	Cross-task master router, for non-short-drama requests	Always, first determine if within scope
24	24-multi-shot-same-scene-camera-base-map.md	Camera base map method for multi-shot same scene, axis locking	Module 2 Storyboarding (multi-shot same scene)
25	25-production-design-blueprint.md	Production design blueprint: top view of scene space, character blocking	Modules 2–3 (complex spatial layout)
26	26-storyboard-master.md	Storyboard master: main line teaching from scene card to coverage	Module 2 Storyboarding (main line)
27	27-storyboard-recipes.md	Storyboard recipe library: rhythm/shot variety/coverage pattern preset	Module 2 Storyboarding (recipe application)
28	28-coherence.md	Coherence check: position/screen direction/prop/A/B group connection	Module 2 Storyboarding (coherence check)
29	29-ai-storyboard-deployment.md	AI adaptation of storyboards: prompt format matching, parameter constraints for executing vintage style, high difficulty	Module 2→3 (conversion to prompts, AI constraints on shot execution feasibility)
30	30-action-scene-causal-grammar.md	Causal grammar: action scene shot-by-shot combination, multi-character interaction	Modules 2–4 (action scenes)
31	31-confrontation-style-lock-long-scene.md	Confrontation scene / style lock / long scene continuous / timestamp rendering	Module 2 Storyboarding (dialogue confrontation)
32	32-user-post-production-sound-music-reference.md	User-posted reference: post-production sound/music style and audio texture specification	Module 5+ (user provides sound reference)
33	33-group-end-technical-panorama.md	Group-end technical panorama: cross-group cohesive state lock, 5-second technical patch chain anchor point	Modules 2–4 (cross-group continuation)
34	34-viral-family-drama-formula.md	Family drama hit formula: reality family/family ethics/social pain point, default track	Module 1 Script (family drama genre)
35	35-script-drafting-modes.md	Dual writing modes: M1 standard screenwriting / M2 novelized narration + orthogonal episode split + tempo gears	Module 1 (developing outline into full script)
36	36-script-doctor.md	Script doctor specialist: outline/episode/scene/complete script diagnosis and revision	Module 1 (user requests script doctor)
37	37-dialogue-expert.md	Dialogue polish specialist: subtext, tone, character voice distinction	Module 1 (user requests dialogue polish)
38	38-action-scene-guidelines.md	Action scene guidelines: combat choreography/weapon handling/impact; for shonen/comic combat, add jump cuts/lazy edits	Modules 1–2 (action genre)
39	39-cross-group-script-notes.md	Cross-group text script notes: out-of-scene character action line: first shot matching log	Module 2 Storyboarding (cross-group)
40	40-extras-direction-and-prop-lock.md	Extras direction: instruction syntax, blocking & crowd forms / prop direction: action-relative direction, asymmetry cadence; single instance: unique attribute lock	Module 2 (extras & props)
41	41-pure-text-prompt-and-ultra-long-dedup.md	Prompt body pure text writing method / ultra-long deduplication & integrity splitting method	Module 3 Prompts (long prompts)
42	42-camera-language-toolbox.md	Camera language toolbox (practical version of No.14 terminology, direct application in storyboards & prompts)	Modules 2–3 (deciding camera movement, shot size, camera placement per scene)
43	43-hook-and-tempo-curve.md	Hook visual method + film tempo curve six-beat structure	Modules 1–2 (hook design & structural control)

---

⚠️ Notes & Important Matters

Skill Boundaries

• Applicable scope: Live-action realistic short drama full pipeline controller (including stylized/comic drama branch)
• Not applicable: Pure editing/compositing/media asset processing (use jianying-editor, byted-mediakit series); commercial ads/voiceovers (use doubao-creative-video); pure 2D dynamic comic final cuts (use ai-drama-video-production)
• Does not contain: Any watermark removal related content
• Does not use: Terminals; entire workflow in chat interface

Dependency / Recommended Module Chain

• Upstream: blockbuster-script-director (script adaptation) / ai-story-vault (topic trend analysis) / video-generation-preferences (user hard constraints)
• Core: This skill (full pipeline master control)
• Downstream: script-asset-master (art asset production) / image-reference (reference image standards) / jimeng-studio·jimeng-director (Jimeng engine deployment) / jianying-editor (post-production editing)

Communication Style Requirements (Interaction Protocol)

• Clear, concise: Avoid redundant polite phrases, express directly and efficiently
• Use the exact format: At each gate, use "### ✅ Phase Complete" + "### Please Choose" + "### ⛔ Paused" format
• Option table format must be complete: Do not omit any of Number/Plan/One-Sentence Highlight/Pros/Cons/Recommendation
• Generate when needed: Do not pre-generate content to save steps, produce on demand
• One short question only: When requirements are unclear, ask only one short question; do not present multi-option menus

Important Notes

• No subtitles, no BGM, TTS disabled (dialogue by video model)
• User character images: One person per role, no duplicates; extra roles pure text description
• Image-to-video only: Must use image-to-video generation throughout, @ the character; absolutely no pure text video generation
• Review before @: All character/scene/prop assets must be reviewed and finalized before @ referencing
• Asset AI direct generation + mandatory review: AI can directly call T2I to generate art assets, but each image/batch must be immediately delivered for user review; finalize @ only after approval; user self-made optional
• Module vertical confirmation: Even within the full pipeline, each module confirms horizontally before proceeding to the next; no default full automatic run without confirmation
• Lossless hard assembly: No re-encoding; no transitions; original quality zero compression; audio-video sync verification
• User self-editing: The delivered piece is a rough cut for user self-editing; this skill does not perform editing, color grading, or sound mixing

---

🔄 Update Log Changelog

Version	Date	Content
1.0	2025-02-15	Initial release: full pipeline controller script→storyboard→prompt→generation→review→assembly→delivery, 33 reference files, phase gate control general principles
1.1	2025-02-20	Added asset AI direct generation capability + mandatory image review gate; changed module advancement from automatic continuous to gate-by-gate pause + options + wait for confirmation; added family drama hit formula No.34; updated 8 action scene guidelines No.38; updated cross-group script notes and extras direction/prop lock
1.2	2025-03-01	Added No.35 dual writing modes, No.36 script doctor specialist, No.37 dialogue polish specialist; added No.33 group-end technical panorama cross-group cohesive state lock; updated No.29 AI storyboard deployment with vintage/nostalgic aspect ratio and high difficulty constraints; refined gate control general principles section 3 and operation method
1.3	2025-03-10	Added No.39 cross-group text script notes, No.40 extras direction and prop lock; merger of action scene guidelines No.38; added 5-second technical patch for cross-group continuation in No.33; added No.41 pure text prompt and ultra-long dedup, No.42 camera language toolbox (practical version of terminology), No.43 hook visual method & full-film tempo curve; updated P1 reading list in single-header collage format; completed/optimized Kling 1.6 ultrawide and style-off/camera standardization; refined wording for mini/subtle camera rule and cascading effect exception; consistently distinguished anchor tail vs. technical patch and A/B mode selection in 33
1.4	2025-03-20	Added "for shots where characters remain still, push/camera movement not required"; camera rule for single center-frame: stunned/silent frames, remove minor camera, stable center composition only; consolidation of 14 terminology practical version into 42; optimized hierarchical P1/P2 reading table with consolidated summary section; enhanced skill boundary/not-in-scope judgment reference for 23