Faceless Explainer Turns a Script Into a Journal-Style Video

Use this skill when a user wants to turn a fact-checked explainer script into a faceless, journal-style knowledge video. The picture is built from voice-over, real screenshots or screen recordings, journal-style cards, and step bars. No real person appears on camera. The skill comes from VaneAi888/vane-faceless-koubo-skill. The original docs, the code, and the listed geometric SVGs are MIT licensed. The whole showcase/reference-only/ folder (the author's real videos, voice-overs, and character images) is not under MIT and must not be reused.

This is a production workflow with human checkpoints, not one-click video generation. The bundled script renders only a minimal example (sequential cards, a subtitle bar, step bars, easing, and one still frame). The author notes that a full cut has not been tested end to end on a new machine.

Triggers

Use it when the user says things like "make a faceless knowledge video", "turn this script into a voice-over video", "journal-style explainer video", or "explain a tool with screenshots and cards". Also use it when the user names this skill directly (Vane's Faceless Explainer Video skill).

Not a good fit:

• A one-line topic where the user expects AI to make and publish the whole video automatically.
• Videos that need a real person on camera, a digital human, or a cloned voice.
• An AI environment that can only chat and cannot read or write files or run commands. It can help plan, but it must not claim it has generated any files.

Hard rules

1. Never skip the five checkpoints. These are voice and script, material list, storyboard, 3-second sample, and full cut. Each one waits for the user's explicit approval before moving on.
2. Do not change the script. Keep the approved script word for word. Ask first before changing wording, pronunciation, or adding lines. If the script changes, update the voice-over and the subtitles together.
3. Do not fabricate. Do not invent back-end screenshots or bills, and do not make up chat logs, success rates, or demo results. Label charts that have no real data as "illustrative". If a visual is missing, list the gap. Do not fill it with unrelated images.
4. Do not replace tools or materials. If a tool, license, or material is missing, say what is missing and let the user decide.
5. Do not publish or upload. Even a finished video must not be published by the agent. Do not send the user's private files anywhere.
6. Mask privacy first. Hide account names, balances, avatars, and anything else unrelated to the content before using a screenshot.
7. Do not reuse the author's material. Do not use videos, audio tracks, people, or screenshots from showcase/reference-only/. Without a real image from the user, do not generate a real person's face from text.

Install

Check first:

ls ~/.claude/skills/vane-faceless-koubo/SKILL.md 2>/dev/null && echo "installed" || echo "not installed"
python3 --version; ffmpeg -version 2>/dev/null | head -1 || echo "ffmpeg missing"

If it is not installed, tell the user where the repository will be cloned and linked. After they agree, run:

git clone https://github.com/VaneAi888/vane-faceless-koubo-skill.git ~/vane-faceless-koubo-skill
mkdir -p ~/.claude/skills
ln -s ~/vane-faceless-koubo-skill ~/.claude/skills/vane-faceless-koubo

Codex uses ~/.codex/skills instead of ~/.claude/skills. Restart the agent after installing.

Running the bundled scripts needs Python 3.10 or later, Pillow (pip install -r requirements.txt), ffmpeg and ffprobe, and a Chinese font file that the user has the right to use. If ffmpeg is missing, tell the user to install it with their machine's package manager. The script does not install it on its own.

Read these files first

Before starting, read README.md, SKILL.md, docs/workflow.md, docs/visual-style.md, and docs/materials.md. Open the others by stage:

Stage	Open
Full order for the first video	docs/first-video.md
Voice-over, speed change, mixing	docs/audio.md
Material handoff and visual sign-off	docs/visual-handoff.md, templates/material-list-zh.csv, templates/visual-signoff.md
Quality check and pre-release check	docs/quality-and-release.md
Publishing copy and review metrics	docs/publishing-and-review.md

Scripts

What to do	Command
Create the project folders (raw, voice, assets, work, out, planning, with templates)	python scripts/init_project.py project
Render the 3-second sample	python scripts/render_example.py --font "<font path>" --seconds 3 --out project/out/sample.mp4
Render the full example (default 1920×1080, 30fps, 8 seconds)	python scripts/render_example.py --font "<font path>" --out project/out/full.mp4
Low-cost test layout	add --width 960 --height 540 --fps 15
Change voice-over speed (does not overwrite the original file)	python scripts/speed_voice.py <source audio> <output> --speed 1.2
Add music to the voice-over and normalize loudness	python scripts/mix_audio.py --video <video> --voice <voice> [--music <music>] --out <output>
Technical check (duration, resolution, loudness)	python scripts/qc.py project/out/full.mp4 --output project/work/qc.json

init_project.py stops with an error if the folder already exists, and it does not overwrite it. qc.py only checks technical settings. It cannot replace a person watching and listening to the whole cut.

Workflow

1. Clarify the task. Topic, audience, platform, aspect ratio (default 16:9, 1920×1080), length, existing script and materials, whether the script can be edited, and the deadline.
2. Verify facts and titles. Every number and quote needs a source record. Give 3 titles and a large cover headline, and let the user pick one.
3. [Checkpoint 1] Voice and script. Deliver the full voice-over and the text. By default you can start with a trial listen at 1.2x speed with pitch unchanged, but do not speed up a real human recording without asking. Listen to the whole track once, paying attention to proper nouns, English words, and numbers.
4. [Checkpoint 2] Material list. Write, line by line, which visual goes with each sentence, using templates/materials.csv. Each line says whether the visual is ready-made material, a screenshot or recording to capture, a card to draw, or something the user must provide. Record the source and license.
5. [Checkpoint 3] Storyboard. Once the materials are ready, fill in templates/storyboard.csv against the voice-over timeline, and give the user an overview.
6. [Checkpoint 4] 3-second sample. Use the final aspect ratio, font, and voice-over to render only the first three seconds. Check that the text is legible on a small screen, the layout is not cramped, and the sound and the picture say the same thing.
7. [Checkpoint 5] Full cut. Edit with one approach that actually runs (Remotion, Manim, Pillow plus ffmpeg, or an editing app). Mix to a target of -14 LUFS with a true peak no higher than -1 dBTP, and report the measured values. Deliver the final MP4, separate cover layouts for 3:4, 4:3, and 16:9, the material sources, and an editable project.
8. Review. About 48 hours after the user publishes, and when they provide real data, write three lines: where viewers drop off, what the evidence is, and the one thing the next episode changes.

Visual rules

Aspect	Standard
Colors	Cream white #F7F4EB, charcoal #1A1A1A, vermilion #D95A4B, navy #2A3B5C, plus white. No warm gold.
Picture elements	Light paper texture, hand-drawn outlines, a three-dot window, step bars, and labels. Decoration must not cover information.
Screen recordings	Center them in a hand-drawn window. Use a step bar at the top to show the current step, and highlight only one point at a time.
Motion	Entrances take 0.3 to 0.4 seconds with about 8% overshoot. Stagger items in the same group by 2 to 3 frames. Use screen shake and text slams only at key moments.
Subtitles	Add them last, keep them in a fixed safe area, and do not shake them with the screen. Check proper nouns and numbers against the voice-over.
Covers	Make them separately from the video. Lay out each of the three aspect ratios on its own. Do not crop directly.

Output format

At each checkpoint, give the user:

1. The file paths for this stage (voice-over, list, storyboard overview, sample, or full cut).
2. The points the user needs to confirm, and the known gaps (missing materials, missing licenses, facts still to verify).
3. The measured technical values (duration, resolution, loudness), with a note that these do not replace a human review.

In one line

Wait for a nod at every checkpoint: voice first, then the material list and storyboard, then the 3-second sample must pass before the full cut. Even a finished cut is not published by the agent.