OmniAiLab AI Director skill
A skill that walks a script to a finished film in stages: it fixes the format, then proposes director, camera and lighting.
- 70 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skill is instructions and references and runs nothing itself.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/omniailab-ai-directorDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skill is instructions and references and runs nothing itself.
This is third-party code. Review the repository files before installing.
What it does
OmniAiLab AI Director takes a film from script analysis to delivery. It first establishes the format: narrative film, short drama, animation, documentary, commercial, title sequence, music video, game trailer or UI motion. If the author cannot explain the task, it asks three questions: where the result will be shown, whether the people on screen are real, and what has to be delivered. The author then picks director, camera, lighting and aesthetics, or hands the choice to the skill. The folder holds 20 director presets, 33 lighting setups with five colour-temperature variants, seven aesthetic recipes and five axes for describing the camera. At the end of each stage the author answers "confirm" or names what to change. Scale is open: from a thirty-second clip to a feature of a hundred scenes, live action and animation alike.
Who it is for. Authors of short films, commercials and music videos who make AI video and want to go from script to delivery with a check at every stage.
Good fit when
- A finished script needs a working order from analysis to delivery.
- The author cannot say what kind of piece is needed, and three questions narrow it to one format.
- The film style is set through a director: one lead and one or two for individual parameters.
- A project of dozens of scenes needs one visual line across the whole film.
Not a fit when
- You want a finished clip from one request with no stages or confirmations.
- You plan to take the single `SKILL.md` file instead of the folder: sub-skills and references sit beside it.
- The task has nothing to do with video or film.
Example request
Run `omniailab-ai-director`: here is the script of a two-minute film; fix the format and propose a director and a lighting setup.Limitations
Copy the whole folder: it holds four sub-skills, references and examples, so `SKILL.md` alone is not enough. The main README and the manual are in Chinese; an English README exists. Codex has no skills directory, so a line pointing at `SKILL.md` goes into the project `AGENTS.md`. There is no fully automatic mode: at every stage the skill waits for the author. A manual describing the stages is published separately, with links in the README.
How to disable. Delete the `omniailab-ai-director` folder from the skills directory; for Codex also remove the line from `AGENTS.md`.
FAQ
How do I install it in Claude Code?
Clone the repository into `~/.claude/skills/omniailab-ai-director` or put the folder under the project `.claude/skills/`, open a new session and type `/omniailab-ai-director`.
How do I connect it to Codex?
Put the folder under the project `skills/omniailab-ai-director/` and add an `AGENTS.md` line telling the agent to read `SKILL.md` and follow its workflow.
Which kinds of work does it cover?
Narrative film, short drama, animation, documentary, commercials, title sequences, music videos, game trailers, UI motion, creator videos and educational video.
How do I update it?
Run `git pull` in the skill folder if you installed by cloning.
Does the skill have a licence?
Yes, MIT: you may use, modify and redistribute it freely.
Discussion
No questions or reviews yet.
Related
SkillsA director skill for AI video: turns a script into a shot list, keyframes and motion prompts for Runway, Kling and Veo
Director skill that drives a video from idea and script to scenes, characters, voices and finished clips
Open Film Skills bundles 22 skills for AI film: screenwriting, direction, assets, storyboards, prompts and post-production.
A director skill for stick-figure videos: from a topic or text it prepares a storyboard proposal and ten-second clip prompts in one of four visual styles
Seven AI video skills: promo film planning, scene and fight prompts, acting direction and generation task reliability checks.
The agent scripts, voices and draws a narrated documentary in code, with one continuous camera over photos, maps and cards.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| OmniAiLab AI Director (this card) | no | 70 | |
| Director Skills | no | 286 | seven AI-video skills for storyboards, scenes and acting; this single skill drives the whole job from format choice to delivery |
| Creative Director | no | 304 | generates ideas and scores campaigns; this one runs a film directing process |
README in short
Version 2.4.0, MIT licence, from the OmniAiLab studio. The folder holds four sub-skills, layers for combat, blocking, animation, documentary and promo work, a visual-language library and seven ready prompt sets. A manual is published on two sites.
README badge
Original SKILL.md
--- name: omniailab-ai-director description: AI导演完整版。影视级全流程 AI 导演 Skill，也是一个会引导创作者做判断的决策系统：开工先判定作品形态（剧情片／短剧／动画片／纪录片／宣传片／片头／MV／游戏 PV／UI 动效／达人视频／教育，说不清时只问三个问题收敛、不硬猜），再让创作者指定导演、摄影机、镜头、布光与美学，或交由本 skill 按作品类型给 2–3 个候选并说明理由；确认后按影视工业标准连续完成 P0 项目接收、P0A 十项剧本解析与全片情绪曲线图、P1 摄影色彩与声音创作基准、P2 角色母板/角色设定板/关键道具母板、P3 场景资产与环境色卡、P4 分镜设计与首帧生成、P5 视频提示词与OmniAiLab参数、P6 在OmniAiLab无限画布中生成剪辑导出成片。媒介不限（真人实拍写实、纯动画／2D卡通／3D游戏引擎渲染风格、三维＋手绘混合），体量不限（30秒短片至上百场次长片），从剧本到成片一站式交付。专业能力不靠形容词堆砌：**导演视界层**以「导演式推理」加载 20 位导演处理空间／机位／光影／时间／潜文本的底层机制，含「反例」纪律（与相邻导演的机制差异）与「主导演＋借用维度」混合规则，选定后产出全片沿用的「导演声明」（见 references/DIRECTOR-VISION SKILL.md 与 examples/director-presets/）；**机位与布光层**提供五维机位坐标（6 机型格式 × 11 镜头性格 × 6 焦段 × 3 景深 × 20 运动）与 33 套布光 × 5 色温变体，并有按作品类型 16 类／场景 22 类／人物 16 类／剧情节点 18 个的调度反查表，用户说不清时主动判断给候选（见 references/CAMERA-LIGHT SKILL.md 与 examples/camera-light/）。内置ACTING表演、EMOTION情绪、LIRA图像（含视觉语言库 D/P/C/R/S 画风体系）、CINEDANCE视频提示词四个子技能，并附打斗层COMBAT（遇对抗场面时接入）、站位层BLOCKING（跨镜站位锁定，多人镜头必用）、美学配方库AESTHETICS（A01–A07 七套成片级配方，P1 主动提案，见 references/AESTHETICS SKILL.md）、提示词教义层（16 段固定块序与十一条铁律、把难的东西放进静帧、物理与连续性怎么写下来，见 references/PROMPT-DOCTRINE SKILL.md）、资产体系层（成对资产、统一命名、状态管理、声音锁定，见 references/ASSET-SYSTEM SKILL.md）、剧本引擎层（五要素戏剧引擎与剧本压力测试，见 references/SCENE-ENGINE SKILL.md）、光学层（视场角锚点、镜头决策树、多镜镜头一致性，见 references/OPTICS SKILL.md）、长片工程层（镜头表卡片、主镜与空间图、压力测试、迭代纪律、法条式锁定，见 references/FEATURE-PRODUCTION SKILL.md）、音乐场景层（先歌后演与唇形同步，见 references/MUSIC-SCENE SKILL.md）、动画／混合媒介层（媒介判定、三维预演与生成的分工、混合美学分权、光决定风格、360° 转身视频、动画原理优先、分幕与先拼全片，见 references/ANIMATION SKILL.md）、纪实／纪录片层（观察式纪实影像工作流：纪实规格、去戏剧化叙事、自然光逻辑、人物跨镜一致性、色卡提取与纪实提示词模板，对话中判定为纪录片项目时启用，见 references/DOCUMENTARY SKILL.md 与 examples/documentary/）、入门引导层（每次对话先定作品形态，见 references/ONBOARDING SKILL.md）、通道选择层（宣传片类项目先判定走**生成通道**还是**代码渲染通道**：五判据＋三个提醒节点＋场景全表＋混合接力，素材核验后进 P4 前锁定，见 references/CHANNEL-SELECT SKILL.md）与宣传片线（14 个商业短片方向的子调度、六步流程与共性硬约束，视频统一在 OmniAiLab 画布中选择 MiniMax H3 生成，见 references/PROMO SKILL.md 与 references/promo/）与**动效工作室层**（代码渲染通道：面向有真实产品界面要展示的产品宣传片／发布视频／功能演示／官网首屏片，用真实页面截图＋2.5D 运镜＋节奏卡点＋电影级音效逐帧渲染出片，自带 157 张动效镜头配方卡、七份规则文档、可跑工程模板与音效素材；启用前先过"适不适合"判定，用途决定形态，视觉语言从产品自身设计系统生长，镜头实现按卡＋参考实现落地，见 references/MOTION-STUDIO SKILL.md、references/motion/（七份规则）与 examples/motion/（157 张镜头卡与工程资产））；并附实战案例库 examples/case-studies/（六个商业级 AI 影视实战案例的成品提示词块与问题解法，含动画案例）、案例配图层 examples/images/（49 张方法示意图 ＋ 关键词反查索引，约 2.4 MB）与宣传片成品样例库 examples/promo-cases/（7 条真实 MiniMax H3 成片的原话输入／成品提示词，按子方向分目录，与子技能一一对应）；**电影艺术知识层**（由电影艺术领域著作系统梳理成的**判断与来源层**：15 个能力域 ＋ 39 张能力卡 ＋ 118 本著作书库索引，覆盖读本到讲戏的作业流程、剧本与叙事（spine／戏剧块／叙述拍／支点）、人物与选角、表演指导、视听语言（关系线与三角形原理／视觉分句／封闭与开放形式）、场面调度与机位、分镜与镜头表、剪辑与蒙太奇判准、声音设计六框架、光线与摄影的表意、美术服装造型、类型与参照、制片与现场、方法与心法（含分镜级拉片法）、类型实操与国别流派；它**只产出判断、依据与清单，提示词写法与执行口径一律归既有各层**，并在 P0A／P1／P4 三个节点主动给出判断，见 references/FILM-ART SKILL.md 与 references/film-art/）。配套操作手册见飞书文档，凡有不清楚先查阅该手册。 version: 2.4.0 author: OmniAiLab developer: Mochiball agent_created: true 开源仓库: https://github.com/tomatoparkdao/omniailab-ai-director 操作手册: https://zcn03zgas1zl.feishu.cn/wiki/P2fhwADXvil24UkNkDCcw1x2nbA --- # AI导演完整版 ## 第 0 步：先定作品形态（每次对话的第一件事） 用户进来说不清要做什么时，**先判定作品形态**，再决定走哪条线。完整引导、话术与三问收敛法见 `references/ONBOARDING SKILL.md`；先看这张速览表： | 形态 | 走哪条线 | 第一步 | |---|---|---| | 剧情短片／长片／短剧 | **主干 P0→P6** | 收剧本（**没给就先问**；不自动载入内置剧本） | | 动画片／混合媒介 | 主干 ＋ `references/ANIMATION SKILL.md` | 先做**媒介判定**，再进主干 | | 纪录片／纪实 | `references/DOCUMENTARY SKILL.md` | 先定**纪实规格八项** | | 宣传片／广告／TVC／品牌片／产品片／发布视频／功能演示／界面动效／电视片 | **先过 `references/CHANNEL-SELECT SKILL.md` 选通道** → 再进 `references/PROMO SKILL.md`（**生成**）或 `references/MOTION-STUDIO SKILL.md`（**代码渲染**） | 先**报通道倾向**，再定**规格** | | 片头／包装／视觉设计／游戏 PV／MV／达人视频／教育 | 宣传片线对应方向（`references/promo/…`） | 先定该方向的必读项 | | UI 动效／产品演示／数据面板（**先分屏还是概念**） | 有真实界面要精确还原 → `references/MOTION-STUDIO SKILL.md`；纯概念 → 宣传片线 `references/promo/ui-motion/SKILL.md` | 先过**通道预检** | | 拿参考片要复刻 | 宣传片线 `.../video-deconstruct/SKILL.md` | 先收参考素材与证据要求 | ### 两条「宣传片」通道怎么分（别选错） 同样叫"宣传片"，走哪条通道**取决于画面从哪来**——这是整支片子最容易选错、也最贵的一个岔路口： | | **生成通道** | **代码渲染通道** | |---|---|---| | 画面从哪来 | 模型按提示词生成 | **真实页面截图／录屏** ＋ 代码逐帧渲染 | | 在哪出片 | OmniAiLab 画布选 **MiniMax H3** | 逐帧渲染，可回交可编辑工程 | | 强项 | 真实的人、实景、氛围、材质与光 | 精确还原界面与排版、真实数据、逐帧可控 | | 对应层 | `references/PROMO SKILL.md` | `references/MOTION-STUDIO SKILL.md` | **判定、时机与话术全部见 `references/CHANNEL-SELECT SKILL.md`**（五个判据 ＋ 三个提醒节点 ＋ 场景全表 ＋ 混合分工）。三条先记住： - **画面主角是真人实景** → 生成通道；**画面主角是界面／数据／文字版式** → 代码渲染通道； - **手里没有可截图的真实界面** → 代码渲染不成立，直接退回生成通道； - **两条通道可以接力**：概念与人物镜头走生成、界面与数据镜头走代码渲染，最后合成在同一时间线。 **报判断的时机**：定形态时**只报倾向、不展开**（此刻还不知道有没有真实界面素材，问了也白问）；**素材核验后、进 P4 出首帧之前必须正式确认一次并锁定通道**；此后不翻案就不再提。