Open-source project
s1dashu/director avatar
s1dashu/director

director's style catalog is commented out of the README while it sits in the repository

Direct complete videos with multi-mode workflows for research, scripts, visual development, shot design, media generation, and delivery.

866 stars113 forksUnknownMIT

At a glance

What is it?
An agent skill that walks a video from one line of premise to clips ready for an editor, across three creation modes. The repository ships its own style definitions and reference voices, but the README hides the catalog and hands media generation to an external CLI.
Who is it for?
director is worth reading if you plan videos with an agent and want a documented process rather than an improvised one, because the mode definitions, the character and voice rules and the confirm before writing prompts are the real content here. Two things to check before relying on it.
Can I use it commercially?
Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
Is it still maintained?
Yes. The repository last received commits 97 days ago.
What is it written in?
GitHub does not report a main language for this repository.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • s1dashu/director866 ★ · +101
Star history
Repository2026-09-172026-11-15
s1dashu/director765866

101 Stars in the last 59 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

This is a skill directory, not an application

director is described as an agent skill for directing and producing complete video. The payload is a skill that an agent reads: SKILL.md at the root, then directories for agents, modes, references, tools, voices and repository assets. There is no package to install, no binary to build and no service to run.

Installation is correspondingly manual. The instruction is to clone or copy the repository into a directory your agent can read as skills. The one exception is Codex, where you can simply tell it to install the director skill from the GitHub URL:

> 从 https://github.com/s1dashu/director 安装 director skill.

That request is written in Chinese and means install the director skill from that repository URL.

In Codex the skill is then invoked explicitly with $director. Every other agent is told to use whatever invocation its own runtime supports, which is an honest admission that the launch command is platform dependent.

The declared primary language of the repository is unknown, which fits: what ships here is Markdown, images and reference material, not code that compiles.

The style catalog is commented out and the reason is in the comment

The README has a Visual Style section that is present in the file and invisible on the page. The comment that opens it says the section is temporarily hidden and will be restored once the style previews have been adjusted.

What is hidden is not small. Inside the comment sits a table of named styles, each with a preview image in repository-assets/style-previews/, a one line description and a link to a style definition file under a mode directory. The visible names include a cinematic 3D animation, a melancholic blue simple line animation, a soft coloured pencil animation, a clean line crayon animation, a dopamine cute 3D animation, an eastern fantasy 3D animation film with separate male and female lead previews, and one extra Storytime style with a white round bodied character.

So the styles exist in the repository, reachable by path, and are not reachable from the page. What a reader sees instead is one sentence telling them they can simply describe the picture they want in their own words. That may be the intended workflow, but the seven definitions sitting in modes/animated-explainer/styles/ and modes/cinematic-drama/styles/ are the more precise option, and the section that would introduce them is switched off.

Two modes carry a demo video and Storytime Animation has none

Each creation mode gets a paragraph describing what it is for. Animated Explainer turns abstract concepts, ideas and knowledge topics into concrete scenes, character actions and visual metaphors. Storytime Animation lets a narrator face the audience to look back at events and also step into a reenactment, for personal experience, the stories of people around you, or first person work explicitly marked as adapted or fictional. Cinematic Drama builds a continuous world with consistent characters, scenes and voices, for turning an existing setting, character design and script into an AI film, an animated short drama or a micro film.

Two of the three are followed by a demo link. Animated Explainer has one, Cinematic Drama has one, and Storytime Animation has nothing between its paragraph and the next heading.

The two links that do exist point at github.com/user-attachments/assets/ URLs rather than at files in repository-assets/. The previews are hosted on GitHub's attachment service instead of living in the repository, so they are not versioned with the code and not covered by the MIT grant the repository applies to its own content.

The generation step belongs to an external CLI, documented in markdown

The technical note is one sentence: media generation is done through the LibTV CLI, and the Jimeng CLI is also supported. Both are documented in the repository, at ./tools/libtv-cli.md and ./tools/jimeng-cli.md.

That is the whole integration surface. There is no install command for either tool, no account or key requirement, no pricing note, no rate limit and no list of what each one can produce, in the part of the documentation available here. The tools/ directory therefore holds documentation about software that is not bundled, which is the difference between a skill that plans a video and a skill that can render one.

This matters when you decide whether the skill is usable as written. Everything up to the shot list is text the agent can reason about inside your session. The final segment, turning the plan into clips, depends on a third party service whose setup the repository assumes rather than describes. The one thing stated plainly is the licence position: the repository's original content is MIT, while third party documentation and external links stay under the terms of their own rights holders.

Three modes, and three different continuity requirements

The character and voice section is where the modes diverge most, and it is the part of the README with real instructions.

For anything that appears across shots, a character image is established first, before video production starts. Dramatic work goes further: main scenes, costumes, props and speaking roles each get a unified reference so that separate clips stay continuous. Voice is either chosen from the built in library, documented at ./references/reference-asset-library.md, or created from the first clip onward for that specific project.

Then the split. Storytime Animation holds a fixed narrator identity, which is what makes a retrospective first person piece coherent. Cinematic Drama designs a separate voice per character. So the same repository asks two different questions about voice depending on the mode, and picking the mode first is what determines how much reference work comes later.

The starting examples reinforce that sequencing. Each is a prompt that asks the agent to confirm the script and the visual direction with you before it proceeds, rather than generating straight through.

Privacy boundaries are a line in a prompt, not a setting

The personal story example asks for three things before any writing happens: the story details, the privacy boundaries, and the narrator's persona.

That is the only place in the repository where privacy appears as a consideration, and it is a request to the agent rather than a mechanism. There is no configuration file in the tree, no environment variable, no flag and no settings screen. The whole configuration surface of this skill is the conversation you have with it, plus whatever the agent runtime already provides.

For a mode built around personal memories that is a deliberate simplification, and it has a practical consequence: the only durable instruction you can give is the one you type. Anyone who later opens the same project, or copies the session, gets whatever was said in the prompt rather than a stored boundary.

The Storytime description also asks authors to mark clearly when a first person work is adapted or fictional. That is a disclosure instruction to the writer, and it is worth reading as the same kind of thing: an expectation the skill places on the output rather than a check it can run.

The repository holds voices, previews and agent definitions alongside the skill

The root is small and legible: .gitignore, LICENSE, README.md, SKILL.md, and six directories. agents/ holds the agent definitions the skill relies on. modes/ holds one directory per creation mode, with the style definitions under each. references/ holds the asset library that the voice section points at. repository-assets/ holds the style preview images named in the commented out section. voices/ holds voice material at the top level, alongside the reference document that describes the library. tools/ holds the two CLI write ups.

That is a well shaped skill repository, and it is worth reading directly rather than through the README, because the README's most informative section is disabled and the two demo videos live outside the tree.

The one thing the layout does not include is anything executable. Combined with the unknown primary language and the absence of GitHub releases, the practical reading is that this is a document set at version zero with no published history. The last push is dated 2026-08-10, which tells you when the text last changed and nothing about whether it has been run end to end by anyone else.

Editorial conclusion

director is worth reading if you plan videos with an agent and want a documented process rather than an improvised one, because the mode definitions, the character and voice rules and the confirm before writing prompts are the real content here. Two things to check before relying on it. First, the media generation step depends on an external CLI that the repository documents but does not install or configure, so confirm your access to that service before investing in a script. Second, the visual style catalog is switched off in the README while the definitions sit in modes/, so read those files directly or you will miss the seven named styles. The privacy language is a prompt rather than a control, so if the material is sensitive, treat what you type as the only boundary you have.

Frequently asked questions

What is the director skill and what does it produce?

It is an agent skill for directing and producing a complete video. You can start from one line of premise, a personal experience or a finished script, and it helps with content structuring, a script or breakdown, visual direction, characters and voice, shot planning, and then generating video clips that are ready for editing.

How do I install the director skill?

Clone or copy the repository into a directory your agent reads as skills. If you use Codex you can instead tell it to install the director skill from https://github.com/s1dashu/director, and then invoke it explicitly with $director. Other agents use whichever skill invocation their own runtime supports.

Which creation modes does director offer?

Three, chosen by the narrative core. Animated Explainer for concepts, mechanisms and knowledge topics, Storytime Animation for personal experience or clearly marked adapted first person stories, and Cinematic Drama for an existing world, character design and script turned into an AI film or short drama.

What does director need to actually generate video?

An external command line tool. Media generation goes through the LibTV CLI, and the Jimeng CLI is also supported, with both described in the repository under tools/. The repository ships documentation for those tools rather than the tools themselves, and no installation or account steps for them are given.

Can I choose my own visual style with director?

Yes, you can describe the picture you want directly in your own words. The repository also holds named style definitions per mode under modes/, with preview images in repository-assets/, although the README section that listed those styles is currently commented out pending style preview adjustments.

What licence covers the director repository?

The repository's own original content is released under the MIT licence. Third party documentation and content reached through external links remain under the terms of their respective rights holders, which covers the CLI write ups and the demo videos hosted outside the repository.

Official sources

  1. Issues
  2. License: MIT
  3. README
  4. s1dashu/director on GitHub
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/s1dashu-director.svg)](https://hysenlabs.com/projects/s1dashu-director)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/s1dashu-director.svg)](https://hysenlabs.com/projects/s1dashu-director)

Was this useful?