s1dashu/director

director – AI‑powered “Director” skill for end‑to‑end video creation

What it is – director is an Agent Skill that lets a language model act as a video director. Starting from a short prompt (a theme, a personal story, or a finished script) the skill guides the model through content planning, script writing, visual style selection, character and voice design, shot‑by‑shot prompting, and finally generates video clips that are ready for editing.

Why it matters – Most AI video tools stop at a single clip or a storyboard. director stitches together the whole production pipeline so the user can collaborate with an LLM to produce a complete, coherent video (animation explainer, personal‑story animation, or short cinematic drama) without manually handling each step.

---

Key capabilities

Mode	Goal	Typical output
Animated Explainer	Turn abstract concepts into visual metaphors and animated explanations.	2‑minute educational video with hand‑drawn or stylized animation.
Storytime Animation	Convert a personal anecdote into a narrated animation that mixes a talking‑head style with reenactments.	Narrative‑driven animation that respects privacy boundaries and keeps a consistent narrator voice.
Cinematic Drama	Produce short AI‑driven films or micro‑movies from a pre‑written script.	Multi‑scene, 15‑second‑chunked footage with consistent characters, locations, and voice actors.

All three modes share a common workflow:

1. Idea & audience definition
2. Research / script drafting
3. Character & visual style selection (choose from built‑in style library or describe a custom look)
4. Voice / sound design (pick from an internal voice library or generate a new voice)
5. Shot‑by‑shot prompt generation
6. Clip generation via the supported CLI tools (LibTV or 即梦)
7. Review & final edit – the skill returns a set of ready‑to‑cut video files.

---

How it works under the hood

• The skill is a collection of prompts and helper scripts that an LLM (e.g., Codex) can call using the $director command.
• Media generation is delegated to external CLI tools:
  • LibTV CLI – a command‑line interface for a text‑to‑video model.
  • 即梦 (Jimeng) CLI – another text‑to‑video backend.
• The skill stores reference assets (character designs, voice presets, style sheets) in the references/ folder, which the LLM can query during a session.
• Users install the skill by cloning the repo into the agent’s skills directory or by issuing a one‑liner in Codex: 从 https://github.com/s1dashu/director 安装 director skill.

---

Getting started

1. Install – clone the repo or use the Codex shortcut shown above.
2. Invoke – in a compatible agent, type $director followed by a natural‑language request, e.g.

   使用 director 创作一支两分钟动画科普视频，主题是"为什么人会拖延"。语气温柔，使用手绘风格，先和我确认讲稿与视觉方向。
   

3. The agent will:
   • ask clarifying questions,
   • propose a script and visual style,
   • generate per‑segment prompts,
   • call the chosen CLI to produce video clips, and
   • return the clips for you to assemble.

---

Who it’s for

• Content creators who want AI assistance for full‑video production rather than isolated clips.
• Educators & marketers needing quick animated explainers.
• Indie filmmakers looking to prototype short dramas with AI‑generated footage.
• Developers of AI agents who need a ready‑made “video director” skill to plug into their bots.

---

License & contributions

• The repository’s original code and assets are released under the MIT License.
• Third‑party references keep their own licenses.
• Contributions are welcomed – star the repo, try the skill, and submit issues or pull requests.

---

Quick links

• [Agent Skill definition – SKILL.md] – formal description of the $director command.
• [LibTV CLI docs] – how the skill talks to the video model.
• [即梦 CLI docs] – alternative backend.
• [Reference asset library] – built‑in character and voice presets.

---

Bottom line – director bundles prompt engineering, asset management, and video‑generation back‑ends into a single, LLM‑callable skill that lets you co‑create complete videos from idea to final clips.