s1dashu/director

director

English · 简体中文

Direct complete videos—from the first idea to production-ready media.

Core Capabilities · Examples · Visual Styles · Installation · Get Started

director is an Agent skill for directing and producing complete videos, from the initial idea and script to shot design, media generation, and delivery.

• Animated Explainer — Explain concepts, ideas, history, or knowledge through clear narration and animated scenes.
• Storytime Animation — Turn first-person experiences into animated stories that combine direct-to-camera narration with reenacted moments.
• Visual Journalism — Cover current affairs, business, industry, and other real-world topics through evidence-led storytelling that combines documentary footage, explanatory animation, maps, charts, and motion graphics.

What It Can Do for You

• Choose the right directing grammar. Match the workflow to the force driving the video: personal experience, concept explanation, real-world evidence, or dramatic action.
• Make complex ideas easy to follow. From topic research and editorial decisions to narration structure, it helps you build a line your audience can understand and wants to keep watching.
• Turn a script into a production-ready directing plan. It breaks the narrative into workable units, directs shots and performances, and produces generation-ready prompts for the selected Mode.
• Keep the entire video consistent. Prompt-defined styles, required character references, and voice anchors reduce character drift, style shifts, and voice inconsistency across clips.
• Go from idea to editable clips. Beyond writing, it can continue through reference planning, clip generation, and task tracking, then hand the generated clips to your editor for final assembly and light cleanup.

Video Examples

Built-in Visual Styles

The Skill currently includes six visual languages validated for Animated Explainer and one built for Storytime Animation. Each Mode owns its styles so users are not offered unvalidated combinations. A style can become cross-Mode only after it succeeds in real productions across those Modes.

These styles are starting points, not templates. The Skill redesigns the setting, characters, and shots around each new topic. You can also provide a custom textual style for the selected Mode; visual style references are not image assets.

Storytime Animation has its own Clean White-Character Storytime Animation: rounded white 2D characters, crisp black outlines, limited flat colors, expressive performance, and environments that are more detailed than the cast.

More Than a Look: Reusable Voices

Alongside the seven prompt-defined visual styles, the Skill includes multiple standardized Chinese and English voices.

You can use one of these voices directly, or create a dedicated voice for the current production from the first clip. The Skill asks you to make the choice—it never silently decides the style or voice for you.

See the complete built-in voice library.

Production-Validated Workflows

Storytime Animation adds first-person story collection, a reusable Storytime-only character library, conversational character co-design, narrator performance, and flexible movement between direct address and reenactment. Animated Explainer retains the production workflow that existed before the Mode architecture.

1. Decide what to say. Research the topic around your audience and target duration, then write a clear, well-structured narration script.
2. Decide how it should look. Choose a built-in or custom prompt-defined style, then create references only for characters who must remain recognizable.
3. Direct the words into scenes. Break the script into balanced segments and design concrete events, multi-shot direction, and generation-ready prompts for each one.
4. Validate before scaling. Produce the first clip, confirm the visual and vocal direction, then lock the voice and required character references before generating the rest.
5. Finish in an editor. Download every generated clip, then manually assemble and lightly trim failed edge frames, tiny end-of-clip audio glitches, pacing, and cut points. Automated assembly remains available only when explicitly requested.

The production rhythm validated so far breaks a 1–5 minute video into 15-second clips. Animated Explainer typically targets roughly 60 Chinese characters or about 32 English words and around five shots. English Storytime targets 30 spoken words, normally 28–32, and usually uses 3–5 shots, with about four as the current stable starting point.

Installation

Clone or copy this repository into a skills directory your Agent can access.

If you use Codex, you can also tell it directly:

Install the director skill from https://github.com/s1dashu/director.

In Codex, skills are explicitly mentioned with a $ prefix, so the invocation is $director. The skill name itself remains director; other Agents should use the invocation convention of their own runtime.

Migrating from animated-voiceover

This project was renamed in place from animated-voiceover to director. GitHub preserves the repository history, stars, issues, and redirects from the old repository URL, but a copied local skill directory does not rename itself.

• Update existing Git remotes to https://github.com/s1dashu/director.git.
• Remove the old local animated-voiceover skill directory after installing director; keeping both can expose two versions of the same workflow to an Agent.
• Replace explicit Codex mentions of $animated-voiceover with $director. For other Agents, use their own skill invocation syntax.

Start with a Single Sentence

Once installed, you can begin with a request like this:

Use the director skill to create a two-minute animated explainer: “Stoicism in Two Minutes.”

Or include more creative direction:

Use the director skill to turn “Why do people procrastinate?” into a 90-second psychology explainer. Use a gentle tone and a hand-drawn style, and confirm the script and visual direction with me first.

The Skill will guide you through the necessary choices. You do not need to understand Seedance prompting, voice anchors, or multimodal asset connections in advance.

Tools and Scope

The officially maintained and production-tested media execution path currently uses LibTV CLI. The underlying methods for narration structure, visual consistency, character references, multi-shot direction, and voice management are not tied to a single platform. After reviewing the latest official documentation for the target platform, they can also be adapted to Higgsfield, Jimeng, and other multimodal generation CLIs.

The current workflow is designed primarily for 1–5 minute videos assembled from multiple 15-second clips. Seedance 2.5 Pro’s 30-second clips have not yet been systematically validated, so they are not presented as a default capability of this release.

For the full execution rules, read SKILL.md. The methods for narration, video prompts, voice references, and cover production are organized under references/.

License

Original content in this repository is released under the MIT License. Third-party documentation and externally linked content remain subject to their respective licenses and rights.

If you want to direct better videos with Agents, try it, share it, and give the project a Star.