Make 3D heads from Claude Code, Codex or Cursor.

The first run, step by step

1. Paste one prompt

   Open your agent in the folder with your photo and paste this. It works in any agent that can run a shell command or fetch a URL; the agents page has a version for each one.

   Prompt · EnglishCopy prompt

   Read https://faceninja.net/skill/SKILL.md and install the FaceNinja skill it describes. Then sign me in with it (it will show me a code to approve) and make a MetaHuman head from ./photo.jpg. Save the GLB and the MetaHuman ZIP in ./faceninja-out.

2. Approve the sign-in code

   The agent shows you a short code and opens faceninja.net/device. You sign in there and approve the code. No API key is pasted into the chat; the skill keeps its key on your computer and never prints it.

3. Confirm the photos are yours to use

   Before the first head, the agent shows you a statement to confirm in your own words: you have the right to use the photos and the consent of each person, they are adults, and you follow the Content and Likeness Policy. The agent never confirms it for you.

4. The agent states the cost first

   It checks your balance and tells you what a head costs and what is left before it spends anything. For more than five heads it stops and asks you to confirm the total.

5. Try the free samples

   With no credits at all, the agent can download the 20 sample heads, so you can test the rest of your pipeline first.

6. One head, then a folder

   For a cast, put each person's photos in their own subfolder and paste the cast prompt. The agent counts the heads, states the total and waits for your yes.

   Prompt · EnglishCopy prompt

   Read https://faceninja.net/skill/SKILL.md and use the FaceNinja skill. Every subfolder of ./cast is one person: make one MetaHuman head per subfolder from the photos inside it. Before you start, tell me the number of heads, the total cost and my balance, and wait for my yes. Save each head's FBX and MetaHuman ZIP in ./faceninja-out/<subfolder name>.

What the skill runs

The skill is one Python file with no packages to install (Python 3.9 or newer). Your agent runs it; you can run it too. This is its own help, unedited:

$ python3 faceninja.py --help
usage: faceninja.py [-h] [--json] [--base BASE] [--version] COMMAND ...

FaceNinja: photos in, MetaHuman head out.

positional arguments:
  COMMAND
    login      sign in by approving a code in the browser
    whoami     who is signed in
    balance    credits left and what a head costs
    make       make one head from 1-15 photos of one person
    status     where a head stands
    download   download a finished head's files
    batch      one head per photo, or per subfolder with --per-person-
               subfolders
    samples    the free sample heads: list them, or --download ID|all (no
               credits)
    list       your heads, newest first
    logout     revoke and forget the saved key
    update     update the skill now (it also updates itself at most hourly)
    guidance   the server's current rules and notices for agents
    attest     the user confirms, once per computer, that they may use the
               photos (asked before the first head)

options:
  -h, --help   show this help message and exit
  --json       one JSON object on stdout, progress on stderr
  --base BASE  API server (default $FACENINJA_API_BASE or
               https://faceninja.net)
  --version    show program's version number and exit

What the agent can and cannot do

• Make heads from 1–15 photos of one person, one head per subfolder, and download GLB, FBX, OBJ, STL, PLY or the MetaHuman ZIP.
• Check your balance, list your heads and download them again.
• Write a pipeline script with idempotency keys and a webhook handler, from llms.txt and the skill.

• Buy credits or turn on automatic top-ups: it sends you to the web app instead.
• Confirm consent or approve a sign-in for you.
• Make heads of children, of public figures, or from photos found online.

Questions about heads from AI agents

01Which agents work?

Claude Code, Codex, Cursor, Gemini CLI and OpenClaw, and any agent that can run a shell command or fetch a URL.

02Do I paste an API key into the chat?

No. You approve a code in your browser, and the skill keeps the key on your computer. It never prints it.

03Can the agent spend without asking?

It tells you the cost and your balance before every job, and stops for your confirmation above five heads.

04Is there an MCP server?

Not today. The skill works through a command-line script that agents run directly.