Blueprint Explainer — a ChatCut skill

Turn one topic into a finished concept-explainer video in a paper-blueprint visual language: cobalt ink on warm paper, editorial serif narration, mono annotations, printed texture, and Motion Graphics cut to the exact frame each word is spoken.

Frames from a 3-minute vertical explainer on the Kakeya conjecture, built end-to-end with this skill.

The look

Every readout on screen is computed from the shape actually being drawn — never scripted separately. When the figure is measured at 0.233, the polygon on screen really has that area.

Graphics carry mechanism; language lives in the caption track. Text on the artboard is limited to numbers and mathematical symbols, so the same render works for any narration language.

What you get

A ChatCut timeline you can keep editing — not a flattened file:

• a beat-by-beat script, checked against sources when the topic is factual;
• continuous voiceover with generated silence trimmed away;
• Motion Graphics synchronized to final spoken-word timing;
• captions (bilingual if you want), background music, and an end card;
• structural, timing, caption, and rendered-pixel verification.

Export is a separate, explicit step.

Requirements

This is a skill, not a standalone tool. It drives the ChatCut plugin, which supplies the timeline, Motion Graphic, voiceover, and music tools. Without that plugin connected, the skill has nothing to call.

• Claude Code or Codex as the agent host
• The ChatCut plugin installed and authenticated (below)
• Internet access when the topic needs fact-checking

Voiceover, music, and export consume ChatCut credits.

---

1 · Install the ChatCut plugin

Claude Code

claude plugin marketplace add https://github.com/ChatCut-Inc/agent-plugin.git#main

claude plugin install chatcut@chatcut-inc

Authenticate:

sh "${CLAUDE_PLUGIN_ROOT}/skills/chatcut-plugin-basics-claude/login-chatcut.sh"

Verify:

claude plugin list

Two things that trip people up:

• The marketplace step clones a large repository and can take several minutes. On a slow or proxied connection, raise the timeout: CLAUDE_CODE_PLUGIN_GIT_TIMEOUT_MS=900000 claude plugin marketplace add ...
• The tools only appear in a new session. Claude Code locks its tool table at session start, so start a fresh conversation after authenticating.

If you previously used a standalone ChatCut MCP server, remove it first so the two don't collide:

claude mcp get chatcut && claude mcp remove chatcut

Codex

Use the Codex CLI bundled with the desktop app — the plugin must be installed there, not in the web version.

codex plugin marketplace add https://github.com/ChatCut-Inc/agent-plugin.git --ref main

codex plugin marketplace list

Note the marketplace name it assigns, then:

codex plugin add chatcut@<MARKETPLACE>

codex mcp login chatcut

Verify:

codex mcp get chatcut --json

Start a new thread afterwards.

---

2 · Install this skill

Claude Code

git clone https://github.com/fun1206-bit/explainer-video-skill.git \
  ~/.claude/skills/blueprint-explainer-video

Codex

git clone https://github.com/fun1206-bit/explainer-video-skill.git \
  ~/.codex/skills/blueprint-explainer-video

Restart or refresh the agent host if the skill isn't picked up automatically.

---

3 · Use it

Use the blueprint-explainer-video skill to make a 60-second vertical explainer
about how GPS finds your location.

用 blueprint-explainer-video skill 做一条 45 秒竖屏视频，通俗解释为什么天空是蓝色的。

Explain compound interest as a 90-second landscape video. Keep the narration
calm and add Chinese captions.

The run pauses at three creative gates:

1. duration and orientation
2. full script and visual beats
3. voice selection

Delegate them explicitly ("pick sensible defaults and keep going") and the skill runs end-to-end without stopping.

---

How narration timing works

Separately generated TTS clips look contiguous on a timeline while still carrying audible leading and trailing silence. Butt them together and the narration breathes wrong. This skill treats every clip as a raw take:

1. generate and transcribe the voiceover;
2. trim the hidden padding around the actual spoken words;
3. repack using audible word-to-word gaps;
4. lock narration timing;
5. build every animation beat from the final placed word frames.

At 30 fps, connected beats target a 6–10 frame audible gap; real concept or act pivots target 12–18 frames.

Because animation is authored after narration is locked, a picture beat can never drift from the word it belongs to.

What "done" means

A run isn't finished until it verifies:

• actual last-spoken to next-first-spoken gaps, not just audio item edges;
• Motion Graphic beats against the final placed word frames;
• captions as rendered on screen, not as stored;
• representative pixels throughout every act;
• no unintended black frames, dead zones, clipping, or hard audio cuts.

Repository layout

.
├── SKILL.md                 complete production workflow
├── references/
│   ├── style.md             visual system + validator-safe MG skeleton
│   └── production.md        timeline, narration, caption, music recipes
└── docs/media/              sample frames from a finished run

License

MIT