A tiny declarative FFmpeg timeline builder in Odin.

The model is deliberately simple:

Odin timeline data
        ↓
filter_complex.txt
        ↓
real ffmpeg process
        ↓
output.mp4

It does not bind libav* and it does not try to wrap every FFmpeg filter. It builds the parts that are annoying to author by hand and leaves an escape hatch with timeline.raw(...).

Put your audio at song.m4a, then:

odin run .

or:

just run

The example recreates the temporal-incident style:

tl.text_file(&timeline, "texts/tuesday.txt", 52, 74,
    tl.Text_Style{
        x = "(w-text_w)/2",
        y = "185",
        size = 42,
        color = "white",
        alpha = 1,
        line_spacing = 16,
    })

tl.countdown(&timeline, 136, 142, []string{"10", "5", "3", "2", "1"})

• init / destroy
• audio_reactive_archive — starts the spectrum + waveform visualizer graph
• text_file — timed text from a file; preferred for arbitrary prose
• text_literal — timed inline text for simple safe strings
• countdown — evenly splits a time range across values
• raw — append any raw FFmpeg filter
• finish — finalize to [vout]
• write_filter — write the generated filtergraph
• render — invoke FFmpeg through core:os.process_exec

FFmpeg drawtext escaping can become unpleasant quickly. textfile= makes the timeline readable and keeps punctuation-heavy copy out of the filtergraph.

The code targets the current Odin core:os.process_exec and core:strings.Builder APIs documented for dev-2026-07.