Download the ready-made Skill
One file with everything in it: the pipeline, the prompts for both styles, and the audio and editing rules. Install it once, then ask Claude to "make a Vox video about…" and it hands you a finished video.
What the Skill is and what exactly it does
A Skill is a small folder with a SKILL.md file that teaches Claude Code a complete workflow. This Skill turns a single request — a topic, or even no topic at all — into a finished explainer video:
- Research and fact-checking: it pulls the numbers and facts from real sources (not from memory) and gives you a list of sources
- Vox-format script: cold open → stakes → evidence → the twist → the close. Each block is 20–24 words = one 10-second clip
- Locked style key: a style reference image is attached to every clip so the whole video comes out looking the same
- Clips + voiceover + editing: all through Higgsfield MCP, and you get one MP4 with burned-in subtitles
Two ready-made styles:
| Style | Look | Best for |
|---|---|---|
| Mixed Media (default) | Flat editorial collage: photo cutouts with paper edges, bold colors, marker circles, abstract charts | Number-driven stories, "why X?", Shorts |
| Paper Diorama (the one in the reel) | Cinematic paper diorama: layered old newspapers, characters with censor bars over the eyes, a single accent color, one continuous FPV camera move | Geopolitics, finance, power, anything you want to feel "cinematic" |
Requirements
- A paid Claude account (Pro / Max / Team) — Claude Code doesn't run on the free plan
- Claude Code installed — commands below
- A Higgsfield account with credits — every clip and voice line burns credits (numbers in step 6)
Windows Install Claude Code — paste into PowerShell:
irm https://claude.ai/install.ps1 | iex
Mac / Linux Paste into Terminal:
curl -fsSL https://claude.ai/install.sh | bash
Installing the Skill
The .skill file is just a regular zip. Extract it into Claude Code's skills folder and you're done — it becomes available in all your projects.
Windows After the file lands in your Downloads folder, paste into PowerShell (one line):
Copy-Item "$HOME\Downloads\vox-motion-graphics.skill" "$env:TEMP\vox.zip" -Force; Expand-Archive "$env:TEMP\vox.zip" "$HOME\.claude\skills" -Force
Mac / Linux Paste into Terminal:
unzip -o ~/Downloads/vox-motion-graphics.skill -d ~/.claude/skills/
Make sure it's installed — open claude and type:
/skills
You should see vox-motion-graphics in the list. If it doesn't show up, close the session and open it again.
Connecting Higgsfield MCP
The Skill generates the images, video and audio through Higgsfield MCP. One command (the official address is from higgsfield.ai/mcp):
You'll need a Higgsfield account — create one hereclaude mcp add --transport http --scope user higgsfield https://mcp.higgsfield.ai/mcp
Then open claude → type /mcp → pick higgsfield → Authenticate (it opens the browser). And confirm:
claude mcp list
Inside the session, ask about your balance — if it comes back with a number, you're all set:
What's my Higgsfield balance?
Run it — the prompts
Open claude in any folder and ask. The Skill runs on its own: it decides the topic, the angle and the voice, shows you a short plan before the first paid generation, and then carries on without stopping.
The simplest request — no topic at all (it looks for a trending topic and picks one):
Run the vox pipeline — a one-minute vertical 9:16 video.
With a specific topic + the Mixed Media style (default):
Make a Vox-style video about "Why did electronic chip prices suddenly jump?" — one minute, 9:16, English narration, subtitles on.
The reel's style — Paper Diorama in dark green / mint / white:
Make a video in the paper-diorama documentary style about [your topic] — one minute, 9:16, model seedance_2_0.
Colors: instead of sepia, use layers of dark forest-green, light mint and white paper — stacked like geological strata, with rough torn paper edges and soft shadows, macro photography with shallow depth of field. The only accent: burnt orange.
Generate a new style key with these colors and attach it to every clip.
To test at minimal cost — just 3 blocks before you spend on a full video:
Try the vox pipeline on topic [X] with just 3 blocks, 720p, gemini_omni — I want to see the style before we make the full video.
Cost and what happens behind the scenes
A one-minute video = 6 blocks × 10 seconds. The free stages: research, script, writing the prompts. The paid ones: clips, audio, subtitles.
| Item | Model | Per unit | One-minute video (6 blocks) |
|---|---|---|---|
| Mixed Media clip | gemini_omni · 720p | 30 credits | 180 |
| Cinematic Diorama clip | seedance_2_0 · 720p | 45 credits | 270 |
| Cinematic Diorama clip | seedance_2_0 · 1080p | 90 credits | 540 |
| Voiceover per block | seed_audio | Depends on the voice | 6 segments |
| Burned-in subtitles | anton font | 0.05 credit | 0.3 |
The numbers come from the Skill file itself (real measurements). Prices change — ask Claude "how much will this generation cost?" before you generate; it requests a cost estimate before the first clip anyway.
The pipeline the Skill follows:
- T · Topic — if you don't give it one, it searches what's trending this week and picks something with a "why/how" question and a surprising number
- R · Research — facts verified from two sources, plus a source list that comes back to you at the end
- 1 · Style key — a style reference image (free for the preset, or one image generation for 16:9)
- 2–3 · Script and prompts — free, in Vox format
- 4 · Clips — N clips in parallel, the style key attached to each one
- 5 · Audio — one documentary narrator for all the blocks, and it checks the length of every segment (9–10.5 seconds)
- 6 · Editing — automatic assembly into one MP4 with subtitles
Troubleshooting + FAQ
| Problem | Fix |
|---|---|
| The Skill doesn't show up in /skills | Make sure the path ended up as ~/.claude/skills/vox-motion-graphics/SKILL.md (not a folder inside a folder), and reopen the session |
| The clip came out realistic / 3D instead of collage | The Skill rejects the suggested preset and redoes the block. If it keeps happening, ask: "Tighten the STYLE and NEGATIVE lines and redo the block" |
| The video came out landscape even though I asked for 9:16 | Known issue — the Skill passes aspect_ratio explicitly. Ask it to redo the clip with "aspect_ratio 9:16" |
| failed / nsfw for no reason | Content filter: political figures in close-up, or explosion imagery. Keep the description a medium shot, or switch that block to gemini_omni |
| The audio lags or rushes | The segment is shorter/longer than 10 seconds. Ask "redo the audio for block N as one continuous sentence" |
| ! Needs authentication | Inside the session: /mcp → higgsfield → Authenticate |