Skip to content

Rendering Video

Overview

The DEMO_VIDEO variable makes a narrated run write video.mp4, frame by frame in a headless browser. The limelight-render command turns the run into its deliverables.

Render A Demo

  1. Render one demo. Select it with -k, because a test that builds no Demo fails in a video run.

    DEMO_MODE=narrate DEMO_VIDEO=1 pytest -k order_approval_demo
    
  2. Make the deliverables.

    limelight-render .demos/order-approval --title "Order Approval" --subtitles --gif
    

A render needs the narrate mode, the pytest plugin, Chromium, and ffmpeg on your PATH.

How It Works

The DEMO_VIDEO variable does not record the screen. The browser runs headless under Chrome's begin-frame control, and the renderer asks the compositor for each frame at a timestamp it chooses. The page runs on virtual time that moves one interval per frame, so each hold, glide, and transition is measured in frames rather than wall time, and a slow machine lengthens the render instead of dropping frames from it.

Quality

The DEMO_VIDEO_QUALITY variable picks one of these.

Quality Output Use
low The frames are 1920x1080 at 24 fps. The level suits a pacing pass, where only the timing is read back.
medium The frames are 1920x1080 at 30 fps. The default level is sharp enough to read the interface.
high The frames are 3840x2160 at 60 fps. The level suits a released demo, which is rendered once.

A high render carries eight times the pixels per second of medium, so measure it on your own page before a long take.

The Command

The limelight-render command takes the directory of a demo and writes its files there. The command always writes chapters.txt and walkthrough.md, and each option adds to them.

Option Description
--title TITLE The option sets the heading the walkthrough opens with.
--subtitles The option writes subtitles.vtt and burns it into render.mp4.
--audio AUDIO The option mixes an audio file into render.mp4.
--gif The option converts video.mp4 to video.gif.

Add A Voiceover

The library generates no speech, so bring a synthesizer or recorded clips.

  1. Run the demo with DEMO_MODE=narrate DEMO_SPEED=turbo to record what it says.
  2. Read the lines with voiceover_cues, and speak each one.
  3. Write holds.json, which maps the text of each line to its length in whole milliseconds.
  4. Render with DEMO_HOLDS=holds.json added to the command above.
  5. Mix the clips with VoiceoverExport, and pass the result to limelight-render with --audio.

A line that is not in the file falls back to the step, so regenerate the file when the script changes. The examples/voiceover_holds.py program measures the holds this way.