Give it the voice track, its word timings and the footage (a talking head, a cut, or
nothing); get back overlay layouts that build on the words: slides beside a presenter,
lists that land point by point, logos, lower thirds, a presenter that shrinks into a
bubble when the slide needs the room. The spec format, the commands, the first look and
the live-spec gate are hn4a-compositor's; this leaf is the layout craft. For titles
that sit inside the footage (type behind people, glitches, typing, reflections), use
hn4a-compositor-in-scene-titles, which writes the same spec.
Inputs
- The voice track (the dialogue master) and its word timings, read with
python-read-word-timingsfrom that exact file. - The footage, and for a cutout a matted copy with a real alpha channel
(
birefnet-matte). - The brand's fonts and colours, and a still of the layout to match if there is one, to lay boxes out against.
Run it
One spec per slide layout (or run of slides), written by a build script
(hn4a-compositor, references/build-script.md), then the commands and the gate in
hn4a-compositor. A complete slide (a title, the presenter shrinking to a bubble,
three points building on the words): references/worked-spec.md.
Cue from word times
Slides change on the words. Take each entrance from the word timings, a little before the word (the client's own slides landed about 0.1 s early): "First," spoken at 23.0 s puts its point at 22.9. When the voice track is re-taken, time the new file and re-cue from it; never nudge numbers by hand.
Layouts that worked
- A presenter beside the slide: the head in a window or cut out on one side, the
copy on the other, the head never covered. When a slide needs the whole frame, the
presenter moves into a round corner bubble (
box,cropandradiuskeyed together, or two layers overlapping 0.5 s). Covering the frame, the take'strim, the bubble's crop on the take's own px, and the helpers that compute them:references/presenter-moves.md. - A list that builds: one layer per point, each on its own word ("First", "Second", "Finally"), earlier points staying on screen.
- A heading with one coloured phrase:
runs, the phrase in the accent colour. - Hand-drawn marks: an arrow or a circle as a
pathlayer withdraw, the shaft over[0, 0.75]of the draw and its head over[0.75, 1]. - Type that tilts into the frame or lies on a surface (a headline on the floor, a
list receding like a credits crawl, a route drawn into a map): a plane group
(
hn4a-compositor-perspective-planes), never a skew or a shrinking box faking it.
Pace the camera
A camera is a keyed group over the layers it moves (hn4a-compositor,
references/spec-format.md, A camera over a group). No one pace is right: a camera
that suited a summary read as "sleepy" in one presenter ad. Choose from the intended
tone, the emphasis and the viewer's reading load: hn4a-compositor-fast-ad-camera
for energetic beats, hn4a-compositor-calm-explainer-camera for inspection and measured
reveals. A held shot is appropriate when movement adds no useful emphasis.
Tips and traps
- A cutout needs a real matte (
birefnet-matte); a colour key leaves a fringe on skin and hair. Or keep the footage in a window or bubble where its own background shows. - A bubble is a square crop of the face with
radius: "50%": measure the face box on a still, never guess it. - Check the longest copy on a still: a heading that fits in one look wraps in a wider font.
Check it
The checks in hn4a-compositor (frame count, stereo, a contact sheet), plus a still at
every entrance, looking for a point that lands before its word or a presenter covered
by copy.


