n0x-content

name: n0x-content description: > Generate brutalist kinetic marketing/promo videos for any project using the n0x-content pipeline. Triggers on "generate a video", "make a promo video", "create a marketing video", "brutalist video", "kinetic video", "n0x-content", "n0x video", or any request to produce video content from project assets. allowed-tools: read, write, edit, bash

n0x-content — Kinetic Video Generation

Generate brutalist, beat-synced short-form videos from a project's assets using the n0x-content pipeline. This skill orchestrates: bootstrap → asset ingestion → LLM-generated cutlist → containerized render.

Engine Location

The n0x-content engine lives at ~/1-projects/n0x-content/. It is an independent project containing the Python renderer (main.py), Podman Containerfile, shell scripts, and project templates. The agent never edits the engine.

Wrapper scripts in ~/dotfiles/scripts/ bridge the engine into pi:

The prompt template is at ~/.pi/agent/prompts/n0x-cutlist.md.

Workflow

Phase 1 — Determine project location

The video project needs a directory. Options, in priority order:

1. User specifies: --output <path> or names a directory
2. Inside current project: If the user is working in a project (CWD), create <cwd>/n0x-video/ as the video project directory
3. Explicit ask: "Where should I create the video project?"

The agent must resolve an absolute path before proceeding to phase 2.

Phase 2 — Bootstrap the scaffold

bash ~/dotfiles/scripts/n0x_bootstrap.sh <absolute-target-dir>

This creates the directory with: audio/, raw_footage/, overlays/, prompts/, output/, config.json, and a README.

After bootstrapping, inspect the current project (CWD) for usable assets:

• Screenshots in the project tree
• A logo or icon (logo.png, icon.png, favicon.png, etc.)
• Any existing video or image files in the project

Copy or symlink discovered assets into the appropriate subdirectories:

• Screenshots/images → raw_footage/
• Logo → overlays/

Check whether audio/ has any files after bootstrapping (it won't, unless the user has pre-placed them).

Phase 3 — Asset ingestion (if needed)

If the project has no audio in audio/:

1. Tell the user: "No audio found. I can fetch royalty-free tracks, or you can drop your own soundtrack into <dir>/audio/."
2. If the user wants auto-ingestion, ask what vibe they want (default: "dark aggressive drill beat") and run:

bash ~/dotfiles/scripts/n0x_ingest.sh <dir> --audio-query "<vibe>"

If raw_footage/ is empty beyond what was copied from the project:

1. Tell the user: "No video footage found. I can fetch public domain films from the Internet Archive, or you can drop clips into <dir>/raw_footage/."
2. If the user wants auto-ingestion:

bash ~/dotfiles/scripts/n0x_ingest.sh <dir> --no-audio --ia-id <identifier>

Use --no-preprocess if the user wants the raw unprocessed look.

Do not auto-ingest without asking. The user should confirm, especially for audio where the vibe matters.

Phase 4 — Generate the cutlist (the agent's core task)

This is where the agent does the creative work.

1. Read the prompt template:

   ~/.pi/agent/prompts/n0x-cutlist.md
   

   Internalize its aesthetic rules, timing discipline, filter/effect vocabulary, and JSON output format.

2. Inventory available assets. List all files in raw_footage/ and overlays/. Note which are videos vs. images.

3. Determine the soundtrack duration. If audio/ has a file, note its length. If not, assume 15–30 seconds (state the assumption to the user).

4. Compose the cutlist. Following the prompt template's rules exactly:

   • Hook phase: 4–6 seconds, slow cuts (1.0–1.5s per segment), Ken Burns, monochrome, tension-building text
   • Drop transition: one or two white flash strobe segments
   • Kinetic phase: rapid cuts (0.3–0.8s per segment), cycling filters, alternating text
   • Use every available asset at least once
   • Text: UPPERCASE, 2–5 words, declarative, mechanical tone
   • No gaps — end of one segment equals start of next

5. Write the cutlist:

   write <project-dir>/prompts/cutlist.json
   

   The content must be valid JSON — an array of segment objects as defined in the prompt template.

6. Report to the user. Summarize: duration, segment count, assets used, text headlines used. Offer to adjust: "Does this look right? I can tweak timing, text, or filter choices before rendering."

Phase 5 — Render

Once the user approves the cutlist:

bash ~/dotfiles/scripts/n0x_render.sh <project-dir>

Optionally pass style overrides:

--resolution 1920x1080     # for widescreen instead of vertical
--font LiberationSans-Bold # font override
--font-size 52             # smaller text
--audio-offset 5           # skip first 5s of audio

The render runs inside a Podman container (auto-built on first use). Report progress and the final output path: <project-dir>/output/render.mp4.

Config Overrides

The default config.json uses the vertical preset (1080×1920, serif font). The user can change the preset or individual settings by editing config.json:

{ "preset": "widescreen" }

Valid presets: vertical, widescreen, square, story, cinematic.

Asset Discovery from the Host Project

When bootstrapping inside an existing project, look for:

• *.png, *.jpg, *.gif — screenshots, diagrams, UI mockups
• logo.*, icon.*, favicon.*, brand.* — branding assets → overlays/
• demo*.mp4, screen*.mp4, *.mov — existing footage
• A README.md — mine it for product name, tagline, and key phrases to use as cutlist text

Always tell the user what you found and ask before including sensitive or unreleased material in the video.

Dependencies

• podman (or docker) — for the containerized render
• ffmpeg, yt-dlp, wget, ia (internetarchive) — for ingestion
• Run ~/1-projects/n0x-content/scripts/install_deps.sh if deps are missing

Verification

• ~/dotfiles/scripts/n0x_build.sh must succeed before render
• prompts/cutlist.json must exist and be valid JSON before render
• audio/ should contain at least one file (warn if not)