# Stickman Video Director: stickman videos from a topic or script (kaomei/stickman-video-director) — skill card for AI agents

> AgentGid skill card. Checked on 2026-11-12 against commit 8658f9e.
> Every fact comes from the pack's own files at that commit. Numbers marked "as reported" are the author's.
> Human page: https://agentgid.com/skills/kaomei/stickman-video-director/

> Note: the pack has changed since this check; re-read its files before relying on details.

## Explain it to a non-programmer
Your AI assistant turns a topic, an article or a script into a stickman video for YouTube or TikTok. In one mode it writes the story and shot-by-shot prompts that you paste into a video generator yourself. In the other it makes a finished video with voice, captions and sound effects on your computer, after you approve the script and a short sample.

## Who it is for
- You run or plan a YouTube channel without showing your face and want animated stickman explainers or stories
- You have a topic, a link or a script and want a story plan, scene prompts or a finished narrated video from it
- You use Codex, Claude Code, Cursor or another assistant that can work with files
- Not for: You want a finished video but your assistant can only chat: the full video needs an assistant that works with files and runs programs
- Not for: You don't want to pay for any outside service: pictures, voice and video clips come from services you connect and pay for yourself

## What it is
- Kind: knowledge, scripts
- License: MIT
- Repository: https://github.com/kaomei/stickman-video-director

## What it does
- One skill, two modes. A · Director plans the story and writes standalone video prompts; B · Autopilot builds a finished MP4
- Director: asks for screen shape, length (in 10-second steps, 60 seconds by default) and style, writes a proposal and stops for approval, then writes one prompt per roughly 10-second clip
- Director scripts follow a five-stage arc: hook, broken assumption, hidden mechanism, main truth, closing question
- Autopilot: script and voice plan, a configuration check, a 20-second sample with sound, then the full video after you approve that sample
- Autopilot renders on your computer with bundled Node scripts and FFmpeg; pictures and voice come from the services you connect
- Optional finishing for both modes: thumbnail, five titles, description and tags, a pinned comment and a pre-publish checklist
- Never cuts: narration and synchronized subtitles in every Autopilot video; your approval of the exact sample before the full render; asking before any paid call

## Skills
| Skill | Call | Changes code? | In plain words |
|---|---|---|---|
| directing-stickman-videos · A Director | plain language ("Use Director Mode / mode A to turn this topic into a stickman video") | no | Writes the story plan and shot-by-shot prompts; you make the clips and edit them yourself |
| directing-stickman-videos · B Autopilot | plain language ("Use Autopilot Mode / mode B to turn this article into a YouTube video") | yes | Makes a finished narrated video with captions and sound effects on your computer |
| directing-stickman-videos · Thumbnail & Publishing Kit | plain language ("Generate the Thumbnail & Publishing Kit for this video"), or yes to the offer after the final video | no | Makes a matching cover and the text you paste into YouTube Studio |

### directing-stickman-videos · A Director
- Asks for: A topic, copy or brief
- Asks for: Screen shape: 16:9 or 9:16
- Asks for: Length in 10-second steps (60 seconds if you don't say)
- Asks for: A style: A1 to A6
- Gives: Director's proposal: title, hook, narration in English with a translation, voice plan, colours, music and a scene table
- Gives: After your approval: one standalone prompt per clip
- Gives: A stitching guide for joining the clips
- Say: "Use Director Mode to turn this topic into a 30-second 16:9 stickman video in the A3 Prehistoric Cave style. Give me the director's treatment first."

### directing-stickman-videos · B Autopilot
- Asks for: A topic, article, post, news link or script
- Asks for: Length, screen shape and a style: B1 to B5
- Asks for: Narration language and voice (English and a default voice if you don't say)
- Asks for: Access to a picture and a voice service
- Gives: A short script and voice plan to approve
- Gives: A 20-second sample with sound
- Gives: The full MP4 with subtitle files (SRT/VTT)
- Gives: A contact sheet, a verification report and the project folder
- Say: "Use Autopilot Mode to turn this article into a 5-minute 16:9 YouTube video in the B1 Minimal Line Art style. Check my setup first, then give me a script and a 20-second preview."

### directing-stickman-videos · Thumbnail & Publishing Kit
- Asks for: Your approved video, or its approved script for drafts
- Asks for: The publishing language
- Gives: One 16:9 thumbnail image (when an image tool is available and its cost approved)
- Gives: Five title options
- Gives: Description, tags and checked timestamps
- Gives: One pinned-comment question
- Gives: A pre-publish checklist with pending items marked
- Say: "Generate the Thumbnail & Publishing Kit for this video."

## Know before installing
- Two modes, two kinds of help: A gives you prompts to paste into a video tool such as Google Flow; only B delivers a finished video file.
- Outside services cost money and need your own accounts: Gemini for pictures and voice, ai33/ElevenLabs for the B4 narrator, Google Flow or Renoise for A6 clips. The assistant asks before each paid call.
- Finished-video mode installs two packages (sharp and a canvas library) into the skill folder and needs Node.js 20+ and FFmpeg on your computer.
- Your project files live in a separate folder, not inside the installed skill.
- Nothing is uploaded or published for you; you copy the cover and text into YouTube Studio yourself.
- Use only avatars and characters you have rights to. The B5 example uses a Spider-Man-inspired presenter, and the pack grants no rights to it.

## Install (official commands)
- Codex (recommended by the author), Claude Code, Cursor or another agent with file and terminal access: The README gives no install command. It asks you to send the repository link to your assistant and ask it to install the skill; the skill folder is skills/directing-stickman-videos. Codex reads agents/openai.yaml for the display name and default prompt. Remove: `delete the installed directing-stickman-videos folder`.
- Chat-only assistants (Director mode only): Director mode works in a normal chat window; Autopilot needs file access and commands.

## What runs on the user's machine
- Code on the machine: yes (B · Autopilot, A6 finishing and the optional caption aligner)
- Network: optional
- Install: No install hook, no background task. Nothing runs until the assistant runs a script.
- Autopilot setup (npm ci): Installs @napi-rs/canvas 0.1.100 and sharp 0.35.5 into the skill folder; this downloads packages from npm.
- node scripts/autopilot.mjs doctor: Checks Node, FFmpeg, FFprobe, fonts and whether GEMINI_API_KEY is set. No network, key value never printed.
- init / validate / preview / approve / render: Reads a declarative project.json and writes the MP4, SRT/VTT, contact sheet, report, shot cache and checkpoint into the project's output/ folder; FFmpeg is called locally. Remote URLs and code in the project file are rejected.
- gemini-media.mjs speech|image (optional): One POST to Google's Gemini API per call with your narration text or image prompt (and reference images only if passed). Refuses to run without --allow-network and GEMINI_API_KEY; writes a .request.json receipt next to the output.
- align_captions.py (optional): Local Whisper word timing; reads the model from ~/.cache/whisper and downloads it only with --allow-model-download.
- B4, B5 and A6 checks: Local checks and renderers (opening-audio check, production ledger, style lock, B5 character renderer, A6 finishing with Pillow). No network calls found in these scripts.
- Director mode (A1–A5): Text only. Nothing runs.

## Settings
- `GEMINI_API_KEY`: Key for the optional Gemini speech and image script; read from the environment, never pasted in chat
- `FFMPEG`: Path to the ffmpeg executable
- `FFPROBE`: Path to the ffprobe executable
- `AUTOPILOT_FONT`: Font file for captions and text

## Examples (from the pack's own files)
- B1 offline sample project: output format defined in skills/directing-stickman-videos/examples/b1/project.json
- A Director proposal, end to end: output format defined in skills/directing-stickman-videos/references/examples.md

## Author's numbers (as reported, not run by AgentGid)
- Hundreds of channels monetize stickman narrative videos under the YouTube Partner Program. Source: kaomei/stickman-video-director/README.en.md

## Steps only the person can do
- Choose and sign in to a picture and voice service (for example create a Gemini API key and set GEMINI_API_KEY) (skills/directing-stickman-videos/references/autopilot-runtime.md) Then: The assistant runs the optional Gemini script or another route you chose, after you approve the cost
- For B4, register or log in at ai33 and approve credit use for the Brian voice (skills/directing-stickman-videos/references/b4-audio-direction.md) Then: The assistant records the narration with that voice
- For A6, sign in to Google Flow in the browser, or install and authorize the Renoise MCP server yourself (skills/directing-stickman-videos/references/amber-noir.md) Then: The assistant generates the clips and assembles the film
- Install Node.js 20+ and FFmpeg if missing (skills/directing-stickman-videos/references/autopilot-runtime.md) Then: The assistant runs npm ci and the doctor check
- Watch and approve the 20-second sample, then the final video (In the chat) Then: The assistant renders the full film, then offers the thumbnail kit

## Mistakes to avoid
- Not: Expect a finished video from Director mode. Instead: Pick Autopilot (B) for a video file; Director gives prompts you paste into a video tool. Why: Director's output is prompts, not generated video
- Not: Run Autopilot in a chat-only assistant. Instead: Use Codex, Claude Code or Cursor with file access and command execution. Why: Autopilot renders locally and needs files and commands
- Not: Paste an API key into the chat. Instead: Put it in an environment variable such as GEMINI_API_KEY. Why: The pack forbids pasted keys
- Not: Record approval and render the full film because the sample encoded fine. Instead: Let the person watch and approve the exact 20-second sample first. Why: Approval binds the reviewed sample and must come from the person
- Not: Treat a set Gemini key or a passing doctor check as proof generation works. Instead: Report login, model access, quota and billing as not yet verified. Why: doctor only checks local tools and key presence

## What leaves the machine
- Approved narration text or an image prompt, plus reference images you pass with --reference → Google Gemini API (generativelanguage.googleapis.com), Only when the optional gemini-media.mjs script runs with --allow-network. Opt out: Use another tool or your own recordings and images; the renderer accepts local files
- Narration script, prompts and reference images → The services you choose (ai33/ElevenLabs for B4, Google Flow or Renoise for A6), Only during generation you approved. Opt out: Choose another provider or supply your own recordings
- A Whisper model download → Your computer, from the Whisper model host, Only with align_captions.py --allow-model-download. Opt out: Supply aligned captions or a cached model

## Lines in the pack aimed at agents (ask the person first)
- "For each new/additional role voice, **actually use Google Search** before casting" (skills/directing-stickman-videos/references/autopilot.md): Web research on the person's story; check with the person before searching and do not send their script or personal details in queries.
- "keep this setup/promotion link out of README files and other-style guidance." (skills/directing-stickman-videos/references/amber-noir.md): The A6 Renoise setup link carries tracking parameters and is a promotion link; tell the person so and let them decide.
- "Run `npm ci` here if dependencies are missing." (skills/directing-stickman-videos/references/autopilot.md): Installs native npm packages into the skill folder; ask the person before installing.

## May I use it like this?
- commercial-app: yes (MIT: keep the copyright and license notice)
- client-work: yes (MIT for the pack; provider-generated media follow each service's terms)
- redistribute-files: conditions (MIT for code and original B1 art; keep the SIL OFL font notices, the Lucide license and the Kevin MacLeod CC BY 4.0 credit for the B5 example; third-party characters are not licensed)

## Which one to pick
- If You want stickman explainers or stories for a faceless channel, as prompts or a finished narrated video: skill:kaomei/stickman-video-director
- If You want a black-canvas narrated explainer from a topic in one Claude Code or Codex skill: skill:Vincentwei1021/anything2explainer
- If You want polished product promo videos built with Remotion: skill:Vincentwei1021/video-shotcraft
- If You want short single-file canvas animations in any style: skill:cth9191/animate

## Works well with
- agent:openai-codex: The author made and tested the showcase videos in Codex
- agent:claude-code: Autopilot runs in agents with file access and commands, Claude Code included
- asset:ffmpeg: The local renderer encodes the MP4 with FFmpeg
- asset:google-veo: A6's verified sample used Veo 3.1 Fast clips in Google Flow

## Works in and needs
- Works in Codex: skills folder (agents/openai.yaml included)
- Works in Claude Code, Cursor, Antigravity: skills folder
- Works in Chat-only assistants: Director mode only
- Needs Node.js 20+: Autopilot renderer
- Needs FFmpeg and FFprobe: Encoding and probing media
- Needs A picture and voice service (e.g. Gemini API key): Autopilot pictures and narration; the bundled adapter needs GEMINI_API_KEY
- Needs Python with openai-whisper (optional): Word timing for captions

## How to tell it works
Install it, then ask for Director mode with a short topic, a 30-second length, 16:9 and style A3. You should get a director's proposal with narration and a scene table, and the assistant should stop and ask you to approve it before writing any clip prompts.

## Safety
The finished-video mode runs bundled Node and Python scripts and installs two npm packages; read them before use. Keep API keys in environment variables. The A6 Renoise route uses a promotion link from the pack.
