ai-visual-director
"AI视觉导演主控路由。用于 /create /avd/create 故事到视频、角色卡、场景图、故事板、分镜图、视频 Prompt、海报、Prompt QC、状态锁定和子技能分发。普通故事默认 /create standard。"
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/jijiutong--ai-visual-director)HTML
<a href="/listings/jijiutong--ai-visual-director"><img src="/api/listings/jijiutong--ai-visual-director/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
These steps run a script from the network that findsafeskills does not scan.
npx ai-visual-director
You can also use the install script or install from source:
curl -fsSL https://raw.githubusercontent.com/jijiutong/ai-visual-director/main/install.sh | sh
git clone https://github.com/jijiutong/ai-visual-director.git cd ai-visual-director sh install.sh
Restart Claude Code, then enter:
/create Two swordsmen face off before a Buddha statue in a rain-soaked temple. The master realizes his disciple has fallen to darkness, and both draw their swords. 15s, 7 shots.
The default is `/create standard`:
Story intake → Shot budget → Character anchor → Scene anchor → Storyboard → Video prompt → QC → Execution checklist
See the [User Guide](./docs/usage.md) for the complete workflow.
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
SKILL.md · Manifest · 135 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
- ---
- name: ai-visual-director
- description: "AI视觉导演主控路由。用于 /create /avd/create 故事到视频、角色卡、场景图、故事板、分镜图、视频 Prompt、海报、Prompt QC、状态锁定和子技能分发。普通故事默认 /create standard。"
- ---
- # AI Visual Director — 主控路由
- 这是给 agent 执行用的主 Skill,不是产品 README。README 负责介绍、安装和能力展示;本文件只定义触发、路由、必读文件和执行约束。
- ## 触发条件
- 当用户提出以下任一需求时,使用本 Skill:
- - 把故事、剧本、小说片段、分镜想法转成可执行视觉方案
- - 生成角色卡、场景图、故事板、分镜图、视频 Prompt、海报或执行包
- - 使用 `/create` `/source` `/storyboard` `/character` `/scene` `/video` `/style` `/poster`
- - 使用 `/avd/*` 命名空间命令
- - 要求进行角色一致性、场景一致性、参考图锚定、视频平台适配或 Prompt QC
- ## 默认行为
- 如果用户只粘贴故事或说“一键生成/直接出/auto”,默认路由到:
- ```text
- /create standard
- ```
- 默认输出状态为 `draft`。只有用户明确确认、锁定或提交时,才写回项目状态:
- ```text
- draft -> locked -> committed
- ```
- ## 命令路由
- | 用户意图 | 路由 | 必读文件 |
- |---|---|---|
- | 一键故事到视频包 | `/create` 或 `/avd/create` | `sub-skills/create/SKILL.md` |
- | 读取 Obsidian/Markdown/粘贴/批处理 | `/source` 或 `/avd/source` | `sub-skills/source/SKILL.md` |
- | 角色外观、表情、服装、DNA、一致性 | `/character` 或 `/avd/character` | `sub-skills/character/SKILL.md` |
- | 场景空间、光照、材质、地理关系 | `/scene` 或 `/avd/scene` | `sub-skills/scene/SKILL.md` |
- | 故事板、分镜图、镜头设计、全案板 | `/storyboard` 或 `/avd/storyboard` | `sub-skills/storyboard/SKILL.md` |
- | 视频 Prompt、首尾帧、平台参数、拆段 | `/video` 或 `/avd/video` | `sub-skills/video/SKILL.md` |
- | 风格浏览、融合、迁移、导演方法参考 | `/style` 或 `/avd/style` | `sub-skills/style/SKILL.md` |
- | 独立海报、封面、营销视觉 | `/poster` 或 `/avd/poster` | `sub-skills/poster/SKILL.md` |
- | 台词、字幕、口型节奏 | `/dialogue` | `engines/dialogue-engine.md` |
- | 环境音、拟音、音乐、混响 | `/sound` | `engines/sound-engine.md` |
- | 锁定、提交、解锁、检查状态 | `/lock` `/commit` `/unlock` `/check` | `engines/state-commit.md` |
- 先读路由目标文件,再按该文件继续读取模板、规则和引擎。不要只凭本文件生成最终资产。
- `/create` 不替代 `/storyboard`、`/character`、`/scene`;它只在一键总编排中按模式调度三条核心资产子链。
- ```text
- /storyboard 核心资产:故事板
- /character 核心资产:角色卡
- /scene 核心资产:场景图
- ```
- ## 主执行链
- `/create` 是默认总编排链,按需调度角色、场景、分镜和视频模块。具体步骤以 `sub-skills/create/SKILL.md` 的执行链路为权威,本文件只列主控边界:
- ```text
- engines/command-gate.md
- -> sub-skills/create/SKILL.md
- -> rules/format-contract.md
- -> rules/prompt-qc.md
- -> rules/final-video-qc.md
- -> engines/state-commit.md
- ```
- ## 必须遵守
- - 输出资产前必须读取对应子 Skill 和模板文件。
- - 格式边界以 `rules/format-contract.md` 为准。
- - 编号体系以 `rules/numbering.md` 为准。
- - 视频负面词和质量检查以 `rules/qc.md`、`rules/final-video-qc.md`、`rules/prompt-qc.md` 为准。
- - 角色一致性以 `rules/character-consistency.md` 为准。
- - 场景一致性以 `rules/scene-consistency.md` 为准。
- - 默认参数以 `rules/one-click-defaults.md` 和 `api-config.template.env` 为准。
- - 用户没有确认时,不写回 `state/` 主状态;允许在输出中生成 draft 状态块或执行清单。
- ## 模板读取规则
- | 输出类型 | 必读模板 |
- |---|---|
- | 角色卡 | `templates/character-sheet.md` |
- | 场景图 | `templates/scene-card.md` |
- | 全案板 | `templates/full-board.md` |
- | 分镜图 | `templates/quick-board.md` |
- | 海报 | `templates/poster.md` |
- | 台词脚本 | `templates/dialogue-script.md` |
- | 声音设计 | `templates/sound-design-sheet.md` |
- 模板是输出结构的权威源。模板内已有的模块数、参数数、版式编号和禁止项,不要在本文件重复维护。
- ## 能力分层
- | 层级 | 内容 | 默认策略 |
- |---|---|---|
- | A 稳定治理层 | command-gate、format-contract、lock-state、prompt-qc、auto-repair | 所有命令必走 |
- | B 资产生产层 | character、scene、storyboard、video、dialogue、sound、poster | 按用户意图调用 |
- | C 导演增强层 | shot-budget、video-director、motion-physics、emotion-curve、color-narrative | project/standard/full 默认可用 |
- | D 探索发散层 | fusion、multi-version、style-migration、director-imitation、series、mood-slider | 用户明确要求才启用 |
- ## 写回策略
- | 状态 | 触发 | 行为 |
- |---|---|---|
- | `draft` | 默认生成 | 只输出草稿、状态块或执行清单,不改 `state/` 主状态文件 |
- | `locked` | 用户确认锁定 | 写入 `state/lock-state.md`,禁止自动覆盖 |
- | `committed` | 用户确认提交 | 写回 `state/variable-registry.md`、`state/asset-map.md`、`state/project-graph.md` 等持久状态 |
- 涉及状态变更时,先读 `engines/state-commit.md` 和相关 `state/*.md`。
- ## 禁止事项
- - 禁止把 README 的安装说明、营销文案、能力展示复制进本文件。
- - 禁止不读子 Skill 或模板,直接凭记忆输出角色卡、场景图、故事板或视频 Prompt。
- - 禁止把全案板和分镜图混为同一种输出;先用 `rules/format-contract.md` 判断 `output_type`。
- - 禁止默认开启 D 类探索能力,除非用户明确要求发散、多版本、融合、迁移或导演模仿。
- - 禁止在用户未确认时覆盖已 locked/committed 的状态。
- - 禁止输出未经过 Prompt QC、格式合同和视频 QC 的最终执行包。
- ## 全局负面清单
- 所有图像/视频 Prompt 必须按需合并 `rules/negative-prompt.md`,并至少包含:
- ```text
- no watermark, no logo, no random large text, no garbled Chinese, no broken faces,
- no duplicated limbs, no messy panels, no low-quality collage, no text overlay,
- no speech bubbles (unless manga format), no cartoon style (unless specified),
- no flat illustration, no marketing poster style
- ```
README.md · README · 220 lines · 1 finding marked
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
- # AI Visual Director
- <p align="center">
- <strong>Turn Stories into Production-Ready Visual Plans</strong>
- </p>
- <p align="center">
- From character and scene anchoring to storyboard design, video prompts, platform adaptation, and quality control.<br>
- A Prompt Production OS for AI films, short dramas, motion comics, and serialized content.
- </p>
- <p align="center">
- <a href="./README.md"><strong>English</strong></a>
- ·
- <a href="./README.zh-CN.md">中文</a>
- ·
- <a href="./docs/usage.md">User Guide</a>
- ·
- <a href="./docs/capabilities.md">Capability White Paper</a>
- ·
- <a href="./docs/system-architecture.md">System Architecture</a>
- </p>
- <p align="center">
- <img alt="License: MIT" src="https://img.shields.io/badge/License-MIT-111827">
- <img alt="Claude Code Skill" src="https://img.shields.io/badge/Claude_Code-Skill-D97757">
- <img alt="Prompt Production OS" src="https://img.shields.io/badge/Prompt-Production_OS-0F766E">
- </p>
- ---
- > A story goes in; character sheets, scene references, storyboards, video prompts, and an execution package come out.
- > This is not an inspiration toy. It is a visual production line with command gates, format contracts, state locking, asset separation, and multidimensional QC.
- ```text
- Story
- → Intake
- → Director Decisions
- → Character & Scene Anchors
- → Storyboard
- → Platform Prompts
- → Quality Control
- → Execution Package
- ```
- ## Why It Exists
- AI image and video models are already powerful, but production is about more than generating a single beautiful image:
- - A character's face, costume, props, and posture drift between shots
- - Scene geometry, lighting direction, and materials lose continuity
- - Presentation boards, marketing art, and video references get mixed together and introduce noise
- - Prompts grow longer after repeated revisions while creative boundaries become less clear
- - Reference-image, first/last-frame, and text rules cannot be reused directly across video platforms
- AI Visual Director encodes the director's workflow as an inspectable constraint system:
- ```text
- Full capability preserved
- → Default authority constrained
- → Exploration explicitly triggered
- → State committed only after confirmation
- → Video assembled, never redesigned
- ```
- ## Core Capabilities
- | Capability | Coverage |
- |---|---|
- | **53 visual styles** | Film, animation, Eastern aesthetics, science fiction, glitch art, and director-method references |
- | **44 layout systems** | Full boards, character boards, scene boards, timelines, HUDs, vertical formats, and keyframe sequences |
- | **140+ camera techniques** | Shot sizes, camera movements, angles, focal lengths, aspect ratios, transitions, and special POVs |
- | **19 numbering systems** | Automatic mapping for style, emotion, color, expression, body language, environment, material, sound, and more |
- | **8 character-consistency methods** | Character sheets, five-angle views, expression sets, costume and weapon details, DNA, and reference anchoring |
- | **7 scene-consistency methods** | All-in-one boards, nine-grids, panoramas, blueprints, surround captures, and widescreen anchors |
- | **5 video platforms** | Platform-specific prompts for Seedance, Runway, Kling, Luma, and Pika |
- | **12 video QC checks** | Character, scene, lighting, props, action, transitions, subtitles, sound, and asset purpose |
- ## Start in 30 Seconds
- ### Install
- ```bash
- npx ai-visual-director
- ```
- You can also use the install script or install from source:
- ```bash
- curl -fsSL https://raw.githubusercontent.com/jijiutong/ai-visual-director/main/install.sh | sh
- Note: Pipes a downloaded script directly into a shell (curl | sh / wget | bash)
- Line 90 on GitHub
- ```
- ```bash
- git clone https://github.com/jijiutong/ai-visual-director.git
- cd ai-visual-director
- sh install.sh
- ```
- Restart Claude Code, then enter:
- ```text
- /create Two swordsmen face off before a Buddha statue in a rain-soaked temple. The master realizes his disciple has fallen to darkness, and both draw their swords. 15s, 7 shots.
- ```
- The default is `/create standard`:
- ```text
- Story intake → Shot budget → Character anchor → Scene anchor
- → Storyboard → Video prompt → QC → Execution checklist
- ```
- See the [User Guide](./docs/usage.md) for the complete workflow.
- ## Choose Your Entry Point
- | What you want to do | Command | Learn more |
- |---|---|---|
- | Generate a complete execution package from a story | `/create` | [One-click orchestration](./sub-skills/create/SKILL.md) |
- | Read from Obsidian, Markdown, or pasted content | `/source` | [Input source system](./sources/README.md) |
- | Lock character appearance and character DNA | `/character` | [Character workflow](./sub-skills/character/SKILL.md) |
- | Lock scene geometry, materials, and lighting | `/scene` | [Scene workflow](./sub-skills/scene/SKILL.md) |
- | Design shots, storyboards, and full boards | `/storyboard` | [Storyboard workflow](./sub-skills/storyboard/SKILL.md) |
- | Assemble multi-platform video prompts | `/video` | [Video workflow](./sub-skills/video/SKILL.md) |
- | Design dialogue, subtitles, and lip-sync rhythm | `/dialogue` | [Dialogue engine](./engines/dialogue-engine.md) |
- | Design ambience, foley, music, and reverb | `/sound` | [Sound engine](./engines/sound-engine.md) |
- | Explore styles, fusion, or director methods | `/style` | [Style workflow](./sub-skills/style/SKILL.md) |
- | Generate a standalone marketing poster | `/poster` | [Poster workflow](./sub-skills/poster/SKILL.md) |
- Governance commands: `/lock` · `/commit` · `/unlock` · `/check` · `/repair`
- ## Three Production Tiers
- | Capability | `fast` | `standard` | `full` |
- |---|:---:|:---:|:---:|
- | Story intake and shot budget | ✓ | ✓ | ✓ |
- | Character and scene anchoring | Inline DNA | Standard templates | Standard templates |
- | Storyboard and video prompts | Condensed | Complete | Complete |
- | Dialogue and sound design | - | On demand | ✓ |
- | Full board and complete QC | - | ✓ | ✓ |
- | Best for | Rapid validation | Production | Full delivery |
- ## How the System Stays Stable
- The system uses a four-layer capability model. Creative capabilities remain available, but each layer has different default authority:
- | Layer | Invocation policy | Responsibility |
- |---|---|---|
- | **Stable governance layer** | Always on | Routing, format contracts, lock state, prompt QC, and automatic repair |
- | **Asset production layer** | Command-driven | Characters, scenes, storyboards, video, dialogue, sound, and posters |
- | **Director enhancement layer** | Project mode | Shot budget, pacing, emotion curves, color narrative, and motion physics |
- | **Exploratory layer** | Explicit trigger | Multiple versions, style fusion, migration, director methods, and series continuation |
- State lifecycle:
- ```text
- draft Default draft; does not pollute primary state
- ↓
- locked Confirmed by the user; cannot be overwritten automatically
- ↓
- committed Persisted to formal project state
- ```
- Learn more: [Capability Boundaries](./docs/capabilities.md) · [State Management](./state/README.md) · [Format Contract](./rules/format-contract.md)
- ## Architecture Overview
- ```mermaid
- flowchart LR
- A[User Story] --> B[Source & Story Intake]
- B --> C[Command Gate & Format Contract]
- C --> D[Director & Asset Planning Engines]
- D --> E[Character / Scene / Storyboard Assets]
- E --> F[State & Consistency Anchors]
- F --> G[Platform Prompt Assembly]
- G --> H[Scoring / QC / Auto Repair]
- H --> I[Execution Package]
- ```
- | Module | Role | Documentation |
- |---|---|---|
- | `SKILL.md` | Main routing, command boundaries, and execution contract | [Main skill](./SKILL.md) |
- | `sources/` | Paste, Markdown, Obsidian, and frontmatter intake | [Input sources](./sources/README.md) |
- | `engines/` | Routing, director decisions, planning, scoring, repair, and packaging | [Engine map](./engines/README.md) |
- | `docs/` | User guide, capability white paper, architecture, and command reference | [Documentation center](./docs/README.md) |
- | `state/` | State registry, asset mappings, shot snapshots, and project dependency graphs | [State management](./state/README.md) |
- | `rules/` | Format, consistency, numbering, quality, and platform constraints | [Quality rules](./rules/qc.md) |
- | `templates/` | Character, scene, storyboard, poster, and sound templates | [Template directory](./templates/README.md) |
- | `knowledge/` | Camera, lighting, composition, material, performance, and sound knowledge bases | [Knowledge base](./knowledge/README.md) |
- | `platforms/` | Image and video platform adaptation and API integration | [Platform adapters](./platforms/README.md) |
- | `sub-skills/` | Independent user-facing command workflows | [Sub-skills](./sub-skills/README.md) |
- | `imitation/` | Director and studio visual-method reference library | [Style references](./imitation/README.md) |
- | `examples/` | Sample projects and reference outputs | [Examples](./examples/) |
- | `projects/` | Persistence for films, series, and multi-segment projects | [Project system](./projects/README.md) |
- | `.agents/` | Agent configuration directory | — |
- | `.claude/` | Claude Code configuration directory | — |
- For the complete data flow and file-placement rules, see [System Architecture](./docs/system-architecture.md).
- ## Documentation Center
- | Document | Best for | What you get |
- |---|---|---|
- | [User Guide](./docs/usage.md) | First-time users and command lookup | Command selection, production tiers, common workflows, and state commits |
- | [Capability White Paper](./docs/capabilities.md) | Understanding system boundaries | Four-layer model, asset purposes, defaults, and invocation conditions |
- | [System Architecture](./docs/system-architecture.md) | Maintainers and contributors | Execution flow, module roles, directory structure, and extension principles |
- | [Main Skill Contract](./SKILL.md) | Skill developers | Main routing, governance rules, module mapping, and global negatives |
- | [Documentation Index](./docs/README.md) | Everyone | Quick navigation from tasks to authoritative documentation |
- ## Platform Coverage
- | Image generation | Video generation |
- |---|---|
- | GPT Image · Midjourney · DALL-E · SDXL / SD3 | Seedance · Runway · Kling |
- | Flux · Ideogram · Tongyi Wanxiang · Recraft | Luma · Pika · ComfyUI / IP-Adapter extensions |
- See [Platform Adaptation](./platforms/README.md) for configuration and limitations.
- ## License
- [MIT](./LICENSE)
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.