Vox Explainer Skill

vox-explainer

A Claude Code skill that turns one topic prompt into a finished Vox-style explainer video. Script, keyframes, animation, voiceover, music, and local assembly — end to end.

**One topic in, finished film out.**

TOPIC → Script → Voiceover → Keyframes → Animation → Music → Assembly
        (Claude)  (xAI TTS)  (Seedream    (Gemini     (MiniMax  (ffmpeg)
                              5.0 Pro)     Omni Flash) Music 2.6)

All generation runs through the [Atlas Cloud](https://atlascloud.ai) unified API — one key covers every model. Assembly happens locally with ffmpeg.

Install

# Claude Code
mkdir -p ~/.claude/skills
git clone https://github.com/YOURUSER/vox-explainer ~/.claude/skills/vox-explainer

Requirements:

• ATLASCLOUD_API_KEY in your environment (free credits on signup)
• ffmpeg + ffprobe with libass
• Python 3.10+ with requests

Use

Open Claude Code and ask:

Make me a 60-second Vox-style explainer about the Tang Dynasty golden age.

Claude reads the skill, writes the script in beats, generates and measures narration first (VO duration drives all timing), locks a visual style with a Seedream anchor frame, animates each keyframe with subtle collage motion, scores it, and assembles a subtitled, mixed, faststart MP4.

What's inside

File	What it covers
SKILL.md	The six-stage pipeline, project structure, orchestration rules
references/vox-style-guide.md	The writing arc and the mixed-media collage visual grammar, with prompt templates
references/atlas-cloud-api.md	Endpoints, model IDs, async polling pattern, cost budgeting
references/ffmpeg-assembly.md	Conform, concat, ASS subtitles, sidechain music ducking, final encode

Design decisions worth stealing

• VO-first timing. Narration is generated and measured before any video exists. Beat durations become the master clock, so clips are conformed to speech instead of speech being squ