Book trailer

Trailer (16:9): final.mp4 · 1920x1080 · 30 fps · 30.00 s · 17.3 MB · -14.0 LUFS / -1.4 dBTP · captions burned (+ captions.srt for YouTube) Teaser (9:16): final-teaser-9x16.mp4 · 1080x1920 · 30 fps · 15.00 s · 11.4 MB · -14.0 LUFS / -1.5 dBTP · captions burned · poster poster-teaser-9x16.jpg

Both are size-capped copies (deliver exports --targets original --max-mb) of the full-quality masters (236 MB and 87 MB: the film look's animated grain does not compress). Neither page was exported as HTML: the brief did not ask for one.

The request#

"Make a 30-second cinematic book trailer for The War of the Worlds using public-domain illustrations and lines from the novel, plus a vertical 15-second teaser."

Mode: quick. The opening line stated the assumptions and made the studio offer once; it was not taken, so the session stayed in quick mode:

Quick mode: a 30 s 16:9 trailer and a 15 s 9:16 teaser, canvas film on Henrique Alvim Corrêa's 1906 drawings and a NASA Mars mosaic, two verbatim lines read by bm_george, an epic-trailer score in C minor. (Want to see concepts first? Say "studio".)

Contract: Mars was watching; they came; nothing we built could stop them: the novel's own words over its 1906 illustrations, ending on the title and "read it free".

The story (trailer)#

Every cut is on a downbeat of the composed score (project/audio/score.beats.json), every line is from Project Gutenberg eBook #36, and every time lives in one cue table (project/cues.js) that the picture, the burn and the mix all read.

The teaser re-lays beats 1 → 3 → 5 for portrait (not a crop): Mars in a 9:16 window with the shorter line "... that this world was being watched" (6.4 s, captioned without a full stop, because the sentence goes on), the Martian's eye from 5.7 s under "was being watched", the drop at 8.6 s to the machine's head and then the whole machine, and the title on its own end hit (11.474 s, whole on frame 345), set in three lines inside the vertical safe box.

The burn inside a canvas film. showtime's shader transitions blend two layers, and a canvas film has one canvas. project/burn.js registers the ridged-burn from the transitions module with the film canvas as the incoming layer and a second canvas as the outgoing one: on each seek inside the 0.8 s window the film draws once more with a flag set (scenes.js then draws the Mars shot at that same time), the pixels are copied to the second canvas, and the film draws its normal frame. Both are pure functions of T, so check still reports the film as deterministic, and the burn works the same in preview, snap and render.

What was changed from the brief, and why (from the researcher's notes):

• No "1898" date line in the arrival beat: in the novel the invasion comes "early in the twentieth century". The beat types the verbatim "Then came the night of the first falling star." instead; 1898 appears only as the book's year on the title card.
• Beat 4 uses "It was the beginning of the rout of civilisation, of the massacre of mankind." The "Ulla, ulla" cry is the Martians', not a human one.
• The full opening line takes 10.7 s, so it runs across beats 1 and 2 (the burn lands on "intelligences").
• The quotes follow the Gutenberg text, which differs from the 1898 first printing ("this world" vs "human affairs", commas); nothing on screen calls them the 1898 text.

Commands, in order#

showtime is skills/showtime/bin/showtime; <t> is showtime-out/wotw-trailer-20260927-153239, <v> is showtime-out/wotw-teaser-20260927-153239, <p> = <t>/project, <q> = <v>/project.

showtime doctor --quick                                            # 21 pass, 0 fail
showtime job init wotw-trailer --mode quick --platform youtube --goal "..." --assumed "..."   # (4 assumptions)
showtime job init wotw-teaser --mode quick --platform shorts --goal "..." --assumed "..."
showtime new film <p> --title "The War of the Worlds" --duration 30

# pictures and fonts (sidecars in assets/media/*.license.json; sha1 of graphic 02 and PIA00003 match Commons/NASA)
showtime assets media search "Alvim Correa War of the Worlds" --source commons
showtime assets media search "Valles Marineris hemisphere" --source nasa
showtime assets media fetch commons:81894077 --project <p>         # graphic 02, 2767x3000
showtime assets media fetch commons:81894106 --project <p>         # graphic 05 (a 1920 px rendition)
showtime assets media fetch commons:81894182 --project <p>         # graphic 15 (a 1920 px rendition)
showtime assets media fetch nasa:PIA00003 --quality orig --project <p>
python <p>/tools/grade.py                                         # one duotone for the three drawings; Mars keeps NASA's colour
showtime assets font Cinzel --weights 400,700,900 --copy-to <p>/fonts
showtime assets font "Special Elite" --weights 400 --copy-to <p>/fonts

# music first: the cuts sit on its downbeats
showtime audio compose --style epic-trailer --key Cm --dur 30 --sections 0:intro,10:build,20:drop,27:outro -o <p>/audio/score.wav
#   end_hit 27.142; then again with --no-sfx (effects placed by hand, below). Round 3 re-composed it (see below)
showtime audio sfx braam --key C --intensity 1.0 --seed 1 -o <p>/audio/sfx/braam-drop.wav   # (+ braam-open, riser --dur 4,
showtime audio sfx keyclick --variants 4 -o <p>/audio/sfx/key.wav                            #  reverse-hit, impact, boom x2)

# voice: two verbatim lines, the second pinned to the drop ({at=20.45})
showtime voice script <p>/narration.md -o <p>/voice                # watch 0.45-11.19, cost 20.45-26.72 (bm_george; round 3: 20.35-25.80)
showtime transcribe <p>/voice/vo.wav --no-events -o <t>/work/vo-check.json   # round trip: every word as written

# picture, first looks
showtime snap <p> --at 0,4,8.2,8.45,8.571,...,29.9 --sheet --cols 5      # 4 rounds of stills
showtime check <p>                                                 # 2 WARN (typewriter line off-frame, 1.5 s hold) -> fixed -> PASS
python <p>/tools/make_mix.py && showtime audio mix <p>/audio/mix.json -o <p>/audio/mix.wav   # 4 rounds (keys masked, braam vs impact, LRA 9.3 -> 6.4)
showtime preview <t>/project --no-open                            # scrubbed in the browser pane: the burn plays in the player
showtime preview <t>/project --stop
showtime render <t>/project --job <t> --preview                   # 720p draft, 45 s
python <p>/tools/quote_srt.py                                      # caption cues at the book's phrase breaks, from the word times
showtime captions <p>/voice/quotes.srt --style cinematic --burn <t>/preview.mp4 -o <t>/work/preview-cap.mp4
#   the cinematic style blurred the letters themselves: fixed in showtime (see Tool issues), then re-checked at 1080p:
showtime render <t>/project --from 0 --to 5 -o <t>/work/seg0.mp4
showtime captions <p>/voice/quotes.srt --style cinematic --burn <t>/work/seg0.mp4 -o <t>/work/seg0-cap2.mp4
python skills/showtime/tests/run_all.py --fast -k footage          # PASS (23 tests)

# teaser: the same scenes.js, its own cues, score, voice and mix
showtime new film <q> --title "The War of the Worlds (teaser)" --aspect 9:16 --duration 15 --job <v>
showtime audio compose --style epic-trailer --key Cm --dur 15 --sections 0:intro,5.7:build,8.6:drop --no-sfx -o <q>/audio/score.wav   # end_hit 11.474
showtime voice script <q>/narration.md -o <q>/voice                # 0.33-6.78 s
showtime check <q>                                                 # 3 WARN (title and credit past x 916) -> fixed -> PASS

# finals, verify, review (three render rounds each; the last round's files ship)
showtime render <t>/project --job <t>                              # final-7.mp4, 240 MB master
showtime captions <p>/voice/quotes.srt --style cinematic --burn <t>/final-7.mp4 -o <t>/final-8.mp4 --srt <t>/final-8.srt
showtime qa <t>/final-8.mp4 --project <p> --platform youtube       # PASS
showtime snap <t>/final-8.mp4 --at 27.133,27.167                   # title whole on frame 815, first frame after end_hit
showtime review-pack <t>/final-2.mp4 --project <p>                 # round 1 (self-review, review/trailer-round-1.md)
showtime review-pack <t>/final-8.mp4 --project <p>                 # round 2 (review/trailer-round-2.md)
showtime render <v>/project --job <v>                              # final-9.mp4, 87 MB
showtime captions <q>/voice/quotes.srt --style cinematic --burn <v>/final-9.mp4 -o <v>/final-10.mp4
showtime qa <v>/final-10.mp4 --project <q> --platform shorts       # PASS
showtime review-pack <v>/final-10.mp4 --project <q>                # review/teaser-round-1.md

# deliver
showtime assets credits <p> --all -o <t>/work/media-credits.txt    # sidecar credit lines -> credits.txt
showtime deliver exports <t>/final-8.mp4 --targets youtube          # 62.3 MB, not committed
showtime deliver exports <t>/final-8.mp4 --targets original --max-mb 18    # this folder's final.mp4 (17.3 MB)
showtime deliver exports <v>/final-10.mp4 --targets shorts          # 23.7 MB, not committed
showtime deliver exports <v>/final-10.mp4 --targets original --max-mb 12   # final-teaser-9x16.mp4 (11.5 MB)
showtime qa <t>/exports/final-8.18mb.mp4 --project <p> --platform youtube  # PASS (and the teaser copy: PASS)

# round 3 (outside critic): the fixes, then the same checks
showtime audio compose --style epic-trailer --key Cm --dur 29.2 --sections 0:intro,10:build,20:drop --no-sfx -o <p>/audio/score.wav
#   end_hit 25.714 (the hit lands ~3 s before the score's end); the mix fades the ring and the wind carries 29.2-30 s
showtime voice script <p>/narration.md -o <p>/voice                # cost {at=20.35 speed=1.2}: 20.37-25.21 s
showtime transcribe <p>/voice/vo.wav --no-events -o <t>/work/vo-check-r3.json   # every word as written
python <p>/tools/quote_srt.py && python <p>/tools/make_mix.py && showtime audio mix <p>/audio/mix.json -o <p>/audio/mix.wav
showtime snap <p> --at 8.471,21.45,22.1,22.8,25.733 --sheet       # burn, the new wide shot, the title frame
showtime check <p>                                                 # PASS
showtime render <t>/project --job <t>                              # final-9.mp4
showtime captions <p>/voice/quotes.srt --style cinematic --text-scale 1.2 --burn <t>/final-9.mp4 -o <t>/final-10.mp4 --srt <t>/final-10.srt
showtime qa <t>/final-10.mp4 --project <p> --platform youtube      # PASS
showtime snap <t>/final-10.mp4 --at 8.471,16.4,21.9,22.757,1.5,28.571 --compare <t>/final-8.mp4   # before|after
showtime render <v>/project --job <v>                              # final-11.mp4 (picture bit-identical; new audio ending)
showtime captions <q>/voice/quotes.srt --style cinematic --burn <v>/final-11.mp4 -o <v>/final-12.mp4
showtime qa <v>/final-12.mp4 --project <q> --platform shorts       # PASS
showtime deliver exports <t>/final-10.mp4 --targets original --max-mb 18   # final.mp4 (17.3 MB)
showtime deliver exports <v>/final-12.mp4 --targets original --max-mb 12   # final-teaser-9x16.mp4 (11.4 MB)
showtime deliver exports <t>/final-10.mp4 --targets youtube; showtime deliver exports <v>/final-12.mp4 --targets shorts
showtime qa <t>/exports/final-10.18mb.mp4 --project <p> --platform youtube # PASS (teaser copy: PASS)

Rebuilding from this folder. The pictures, fonts, music, effects and voice are not committed (about 31 MB, all reproducible). python project/tools/restore.py <path to showtime> (and the same in project-teaser/) re-runs the fetch, grade, font, compose, sfx and voice commands above; tested on the teaser: the score came back byte-identical and check passed. Then showtime render project.

Features shown#

About the crew. This session had no sub-agent tool, so the director did the scriptwriter and sound-designer passes from their briefs and wrote them up in crew/, and answered the critic's CRITIC.md as self-reviews (review/). The researcher's verified facts (quotes checked against the Gutenberg text and three 1898 printings, licences checked on Commons, NASA and JPL) were prepared before the session and are the only source of claims. A person should still listen to both cuts and take a second look.

QA#

Review (two self-reviews, then an outside critic)#

Trailer, round 1 (on the first captioned final): ship after fixes. Fixed in round 2: the capsized-steamer shot came back dimly for six frames in the dip's second half (the title card had no opaque ground); the title faded in over two frames instead of landing whole on the hit; the faint machine behind the title showed its vertical edges. The call to action now enters 0.14 s earlier (2.58 s on screen): with the title on end_hit in a 30 s cut it cannot hold longer, and the last frame is the card (round 3 showed it could). Round 2: ship. Teaser, round 1: ship.

Review round 3 (outside critic, on the shipped files): trailer "ship after fixes", teaser "ship". It caught what both self-reviews had passed: the wide shot at 21.43-22.87 s was pillarboxed (58 % of the frame near-black on the drop). It also showed the 2.58 s call to action was not a limit of the brief, only of where the composer put the final hit. Fixed: a full-frame pull-out on graphic 15; the score re-composed 29.2 s long so the hit and title land at 25.714 s and the call to action holds 4.0 s (the second quote starts 0.1 s earlier and is read a little faster). Polish also fixed: the burn's square-cell grid, captions 20 % larger, a 32 px credit line, the lid shot lifted, and an ending that decays. Details and measurements: review/trailer-round-3.md. The teaser's picture did not change. Only its audio ending did. Nobody has listened yet: a person should.

Sources and licences#

Not claimed anywhere: a new edition, film or release date; that the quotes are the 1898 text; worldwide public domain; that the drawings illustrated the English edition; that the Mars image is true colour.

Tool issues found (logged for the maintainers)#

• captions --style cinematic (and minimal) blurred the glyphs themselves: with no border, libass applies \blur to the fill. Fixed in lib/st/footage/captions.py (a thin dark border takes the blur; letters stay sharp in a soft halo); footage tests pass.
• captions --burn inside a job does not make the burned file the job's latest final, so qa <job> and review-pack <job> pick the uncaptioned render; both were run on the burned file by path.
• review-pack reads CUE keys only at the start of a line, so acts named by CUE.arrive/CUE.title that shared a line with other keys were dropped (round 1 showed 3 of 5 acts); one key per line fixed it.
• The ridged-burn pre-heat glow showed small square blocks (the shader's value-noise grid). Fixed in round 3 (runtime/transitions/shaders.js: the ridge noise is turned, lightly warped and turned per octave).
• audio compose --sections ...,27:outro silently drops a marker that falls inside the final hit's ring-out, and there is no way to ask for the hit's time: it lands on the grid about the style's tail (3 s for epic-trailer) before --dur. Round 3 worked around it by composing 29.2 s and letting the mix carry the last 0.8 s.
• captions had no text-size control: added --text-scale K in round 3 (footage tests pass, with a new test).
• snap <project> --at t showed the frame before t when t rounded to six decimals fell just under a frame boundary (25.733 s showed frame 771, the render has the title on 772). Fixed in scripts/snap.mjs (it seeks k / fps exactly); new test test_08d_snap_project_lands_on_the_frame.

The project files on GitHub