pixi-effects

For AI

Everything an AI needs to make a video with this library and check it, and nothing it has to read twice.

01

Paste this to an AI

the sentencetap an ending to swap it, then copy

Use https://github.com/yjmtmtk/pixi-effects to make a video: a 10-second title for a bakery, warm cream colours, a soft chime at the end

Endings
02

What it reads

  • AGENTS.md: the first screen of the repository. It tells an AI with a shell which path to take and what not to read.
  • skills/pixi-effects/SKILL.md: the workflow and the rules that cause most failures. Install it for an agent with npx skills add yjmtmtk/pixi-effects.
  • llms.txt (the index) and llms-full.txt (all the documents in one file).
  • ai/CHAT.md: for an AI in a browser chat with no shell and no files. One HTML file; every warning shows in a red box to paste back.
  • Ask the docs instead of reading them: npx pixi-effects help <word> prints the sections about a word; npx pixi-effects recipe [id] lists the tested recipes and prints one with its code.
03

What it runs

  • npx pixi-effects check page.html: library warnings (each says what to change), layout and sound problems, a real export decoded again, and a contact sheet picture to look at.
  • npx pixi-effects render page.html -o out.mp4: export, whole or a range (--range, --scene).
  • npx pixi-effects view page.html: the page with a zoomable timeline under it.
  • probe, frames, trim, sound, concat, audio: look into and cut a finished file, with no ffmpeg.
04

The tested recipes

Each one is built and linted by the test suite. Print one with npx pixi-effects recipe <id>.

  • slam-words
  • lower-third
  • spring-pop
  • rewind-and-replay
  • scene-zoom
  • parallax-strips
  • canvas-trail
  • canvas-rings
  • canvas-pixel-sprite
  • canvas-life
  • canvas-text-mask
  • canvas-text-on-path
  • canvas-bars
  • marquee
  • count-up
  • bar-chart
  • depth-title
  • camera-orbit
  • depth-cards
  • scene-flood
  • slideshow
  • slideshow-cuts
  • music-lofi
  • sfx-minimal
  • sfx-cues
  • progress-ring
  • draw-on-and-type
  • split-text
  • stagger
  • animate-text
  • vertical-text
  • follow-path
  • morph
  • audio-react
  • deck
  • camera-fly-through
  • rack-focus
  • pulse
  • particles
  • particles-snow
  • null-orbit
  • light-leak
  • shared-matte
  • luma-wipe
  • spot-room
  • fog-depth
  • shader-plasma
  • warp-water
  • wiggle
  • gradient-background
  • gradient-shift
  • film-grain
  • subtitles
05

Honest notes

  • The AI has to be able to read a web page: it fetches the repository, and in a chat, ai/CHAT.md.
  • In a chat with no preview window you save the file it gives you and open it in a browser.
  • It is a young library, and the first try may need a couple of rounds: if a red box appears, paste back what it says; each message names what to change.
  • The check can tell the AI that nothing is broken, not that the video is good. That judgement is still yours.