Travel slideshow

Video: final.mp4 · 1920x1080 · 30 fps · 45.00 s · 18.9 MB · -14.0 LUFS / -1.4 dBTP (a size-capped copy of the 153 MB CRF-16 master final-5.mp4, deliver exports --targets original --max-mb 20) · Reels: final-9x16.mp4 · 1080x1920 · 45.00 s · 18.9 MB · -14.0 LUFS / -1.4 dBTP, a native 9:16 re-layout rendered from the same page, not a crop (poster-9x16.jpg) · Showcase loop: exports/loop-small.webp (480x270, 12 fps, 6 s, 0.64 MB) with a exports/loop-small.gif fallback (336x190, 10 fps, 1.40 MB), the whip run at 19.9-25.9 s.

The request#

"Make a 45-second travel slideshow of an Iceland ring-road trip from these CC0 photos, cut to the music, with a little map showing where each place is. Also a vertical version for Reels."

Mode: quick. No questions were asked. The opening line stated the assumptions:

Quick mode: 45 s at 16:9, plus a native 9:16 re-layout for Reels, 30 fps, no voice-over. Eight CC0 photos from Wikimedia Commons, one per stop, starting on a downbeat of "Somewhere Sunny (ver 2)" by Kevin MacLeod (CC BY 4.0; the credit ships in the end card, credits.txt and share.txt). A small map inset follows the route. Wind under the map, water under the waterfalls.

Contract: a loop around Iceland in eight stops, each photo landing on a downbeat, with a map that always says where we are.

What the license audit changed in the brief#

The researcher's audit came before the build, and the film follows it:

• The route runs counterclockwise, not clockwise. Reykjavík → the south coast → Jökulsárlón → Goðafoss (north) → Kirkjufell (west) is anticlockwise with north up. The film keeps that order and uses no direction word on screen or in share.txt; this README names it only to record the correction.
• Three photos were replaced. The brief's Reykjavík photo is a black-and-white clock face shot from inside the tower. Its Vík photo is a close-up of sand. Its "Jökulsárlón" photo is an ice cave, not the lagoon. The replacements are the church exterior (F. Stöhr), the Reynisdrangar sea stacks off Vík (A. Spratt) and icebergs in Jökulsárlón (J. Bishop). All are CC0: they were on Unsplash before the 5 June 2017 cutoff, which the researcher checked in the Wayback Machine.
• No claims the sources do not support. There is no road length, no "every stop is on Route 1" (Þingvellir and Kirkjufell are detours), no "back where we started" over Kirkjufell, no "our trip", and no "chorus". The route line is schematic: it follows the researcher's via points so that it hugs the coast, and it is never called the road.
• On-screen facts come from the audit's allowed list only: "Hallgrímskirkja", "Rift valley on the Mid-Atlantic Ridge", "You can walk behind it", "Reynisdrangar", "Glacial lagoon, Vatnajökull National Park", "On the Skjálfandafljót river" and "“Church Mountain”, 463 m". Every place name keeps its Icelandic letters (Þ ð á í ó ö). The display font (Instrument Serif) has them.

The story, shot by shot#

The bed is 106.6 bpm, so one bar is 2.25 s. Each scene starts on a bar downbeat of the track's own analysis (audio beats, rhythmic: true, pacing: beat_cut), floored to the frame. That is 0-33 ms before the beat, within 1 frame of it (list below). Every transition is centred on its cut.

Scene starts vs. downbeats (the bed's grid from somewhere-sunny-ver-2.beats.json; the music under them is the unedited track from 0 to 38.52 s):

Map marker vs. photo: each marker is at the place's Wikipedia coordinates (research sheet, section 2). Where a photo has camera GPS, it agrees to within about 2 km. The inset lights stop N exactly while photo N is on screen.

The 9:16 cut#

The same index.html renders both. project/tools/make_vertical.py writes project-916/ (same page, media and mix; showtime.json at 1080x1920). The page switches layout in @container (max-aspect-ratio: 5/6):

• The map inset becomes a card in the top band, inside the Reels-safe box (x 64-956, y 220-1440).
• The labels sit in the centre band, and the type follows the 2.2 % rule for tall frames (43 px and up).
• Each photo is re-framed with data-pos-tall (for example Vík at 31 %, so the big stack fills the frame).
• The three map cameras compute their zoom from the element size, so Iceland fills the same share in both aspects.
• The end card drops the photographer names in 9:16 (they are in share.txt and credits.txt) so that it stays above the platform UI.

Commands, in order#

showtime is skills/showtime/bin/showtime. <job> is showtime-out/iceland-ring-road-20260927-162623 and <p> is <job>/project. The photo originals and the render work folders are not shipped. The project has web-size copies (2160 px tall).

showtime doctor --quick                                         # 21 pass, 0 fail
showtime job init iceland-ring-road --mode quick --platform youtube --goal "Make a 45-second travel slideshow ..." \
  --assumed "45 s 16:9 master + native 9:16 re-layout (reels); 30 fps; no voice" \
  --assumed "8 CC0 Commons photos per the license audit (3 of the brief's replaced); order counterclockwise; no direction word" \
  --assumed "music: library 'Somewhere Sunny (ver 2)' Kevin MacLeod CC BY 4.0; scene starts on downbeats; CC0 wind/water"

# photos: license-gated fetch with .license.json sidecars
showtime assets media fetch commons:61668982 -o photos/       # Reykjavík; also 34705109 (Þingvellir), 61799729 (Skógafoss)
#   the other five hit HTTP 429 on upload.wikimedia.org originals (see Iterations 1): their 3840 px (Seljalandsfoss:
#   1920 px) Commons thumbnails were put into showtime's media cache, then
showtime assets media fetch commons:121381245 -o photos/      # ... 47696382, 62221230, 86572139, 61894908: sidecars written
showtime assets sheet photos -o <job>/work/sheet.jpg          # 8 items, none low-res; order + focus points read from it
#   resized to 2160 px tall into <p>/media/01-reykjavik.jpg ... 08-kirkjufell.jpg (sidecars copied, "derived" noted)

# music
showtime audio lib info incompetech-somewhere-sunny-ver-2     # 114.73 s, 106 bpm, key C, CC-BY-4.0 + the exact credit
showtime audio beats audio/sunny.opus                         # 106.62 bpm, confidence 1.0, rhythmic, beat_cut; surge 24.0 s
showtime audio fit audio/sunny.opus --dur 45 -o audio/bed45.wav             # rejected: ends on bar 19, a G chord
showtime audio fit audio/sunny.opus --dur 45 --from 2.28 -o audio/bed45b.wav # rejected: ends on an F chord
#   -> <p>/audio/mix.json: bed-a = the track 0-38.522 s; bed-b = the song's own ending (Dm-G-C) spliced in at that
#      downbeat (track 104.165 s), both by library id so the CC-BY credit is automatic; wind + water ambience; 3 whooshes
showtime audio mix <p>/audio/mix.json -o <job>/work/mix.wav  # -14.0 LUFS, -1.1 dBTP; whoosh gains raised twice (masked)
showtime audio beats <job>/work/mix2.wav                     # the grid runs on unbroken through the splice (38.52 / 40.77 / 43.07)

# project and first look
showtime new dom <p> --title "Ring Road" --duration 45       # page then rewritten (index.html)
showtime check <p>                                           # 4 rounds, see Iterations; final: PASS, 1 WARN (explained below)
showtime snap <p> --at 0,1.9,2.4,3.0,... --sheet             # 5 targeted sheets across both aspects
python <p>/tools/make_vertical.py                            # -> <job>/project-916 (1080x1920)
showtime check <job>/project-916                             # PASS, 1 WARN (the same)
showtime job note <job> --stage plan ... ; showtime job note <job> --stage first-look ...

# final, verify, review
showtime render <p> --job <job>                              # final.mp4 (6m28s); final-2, final-3 after the fixes; final-4 after review round 2 (4m54s); final-5 after round 3 (5m14s)
showtime qa <job>                                            # final-3: PASS, 0 fail, 0 warn, 0 note; final-4 and final-5: PASS, 1 note (end-card hold)
showtime review-pack <job>                                   # read as a self-review (no sub-agent tool in this session)
showtime render <job>/project-916 -o vertical/ring-road-9x16-2.mp4   # round 2 (the first was ring-road-9x16.mp4)
showtime qa vertical/ring-road-9x16.mp4 --project <job>/project-916 --platform reels   # PASS, 1 note (end-card hold)

# deliver
showtime deliver exports <job> --targets original --max-mb 20             # final-5.20mb.mp4 -> final.mp4 (18.9 MB)
showtime deliver exports vertical/ring-road-9x16-2.mp4 --targets reels --max-mb 20   # -> final-9x16.mp4 (18.9 MB)
showtime deliver exports <job> --targets webp-small,gif-small --from 19.9 --to 25.9  # loops (0.64 MB / 1.40 MB)
showtime qa <job>/exports/final-5.20mb.mp4 --project <p> --platform youtube          # PASS (1 note)
showtime qa vertical/exports/ring-road-9x16-2.reels-20mb.mp4 --project <job>/project-916 --platform reels   # PASS (1 note)
showtime assets credits <p> --all -o credits-all.txt        # photo lines for credits.txt

To re-render: showtime render project (16:9), or python project/tools/make_vertical.py and then showtime render project-916. The music comes from the showtime library (core tier) by id, so it is not in the folder.

Features shown#

assets media fetch (Commons, CC0 gate, sidecars) · assets sheet · assets credits · ken-burns (linear, from/to per shot, object-position + data-pos-tall per aspect) · world-map (Mercator, src = countries-50m.json, camera fly and push, highlight, markers, multi-segment routes along via points; three instances: full-frame open, inset, end) · page code on ST.onSeek (visited markers, the final-chord glow) · library music by id with automatic CC-BY credit (mix report → credits.txt) · audio beats · audio fit (run; its trims were rejected, see Iterations 2) · a mix-level splice to the song's own ending · library CC0 ambience (wind, water) · synth whooshes · beat-synced scene starts · WebGL light-leak (a = strength 0.5) · crossfade · WebGL whip-blur · CSS whip-pan · blur-dissolve · native 9:16 re-layout (@container) · check · snap · render --job · qa --platform reels · review-pack · deliver exports (youtube, reels, original --max-mb, webp-small + gif-small loops) · researcher license audit and the critic (self-review).

QA#

• 16:9 final-5.mp4 (the master, after review round 3): PASS (0 fail, 0 warn, 1 note). -14.1 LUFS, -1.1 dBTP. credits: final-5.credits.txt lists all 1 required attribution(s). The note: the end card is a still hold of about 3.8 s (from 41.23 s). It is the credit card, and it needs the time. Shipped copy final.mp4 (20 MB cap): PASS, -14.0 LUFS, -1.4 dBTP, the same note.
• 9:16 ring-road-9x16-2.mp4, --platform reels: PASS (0 fail, 0 warn, 0 note). -14.1 LUFS, -1.1 dBTP. Shipped copy (20 MB cap): PASS, -14.0 LUFS, -1.4 dBTP, 1 note (the same end-card hold).
• check on both aspects: PASS with one WARN each. "Reykjavík" is reported as 17 px. It is the map's SVG label, and the component scales it by 2x (on screen it is about 34 px in 16:9), but check reads the CSS font size without the SVG transform (logged as friction).

Iterations that mattered#

1. Commons rate limit. Three originals downloaded, then upload.wikimedia.org returned HTTP 429 for every other original (retries over ~10 min did not help). Wikimedia's message asks for standard thumbnail sizes. The five files were fetched as 3840 px thumbnails (Seljalandsfoss as 1920 px, since 2560 is not a standard step), placed in the media cache, and assets media fetch then wrote their sidecars. Each sidecar notes the rendition.
2. The ending. audio fit --dur 45 trims after a downbeat with a short decay. From the top that is bar 19, a G chord (the dominant). With --from 2.28 it is an F chord. Neither resolves. The track's real ending is Dm-G-C with the final chord at 108.72 s, so the mix plays the track to the bar-17 downbeat (38.522 s) and splices the ending in at its Dm downbeat (104.165 s) with a 50 ms crossfade. The beat grid of the mix runs on unbroken through the splice. The picture then used it: the end scene starts on the splice, the title lands on the G, and the line glows on the C.
3. Credits went missing (qa caught it). The first mix used a copy of the track file ("file"), so neither the mix nor qa knew it was CC-BY, and qa said "no attribution required". Switching the two bed tracks to "lib": "incompetech-somewhere-sunny-ver-2" gave identical audio (max sample difference 0.0) plus the auto credit. The rebuilt review pack flagged the round-1 file with missing_credits, and final-2 onward ship credits.txt.
4. Overlays vanished inside every transition window (review-pack cuts.jpg). The text and the inset map are overlay clips after the scenes, so shader transitions never smear them. But inside a window the transition engine gives the scenes z-index 1-3 and the WebGL canvas 4, so overlays without a z-index were painted underneath. The inset blinked out for 0.45-0.8 s at every cut. .ov { z-index: 10 } fixed it (final-2). check and qa do not see this.
5. check's text audit could not see the overlays. With pointer-events: none on the overlay layers, check measured contrast for only 1 text element (its hit test goes through them). With it removed, check found real contrast FAILs: the amber stop numbers were 1.7:1 on bright sky. They now sit on a small dark pill, and a bottom gradient plus a soft local scrim sit under the labels. The scrim fades in with the words (final-3), because round 1's cuts.jpg showed it popping on in one frame.
6. Layout rounds. In the 16:9 end card, the map moved right and up to clear the credits. The Reykjavík label is anchored left of its dot. In the tall layout, "Ring Road" overlapped the credits, and the credit type was under 43 px. The inset card shrank from the full width to a 50 %-wide card so the church spire and Kirkjufell's peak stay visible. The Jökulsárlón label got 0.18 s more (check: its 61 characters need about 3.9 s).

Review (critic, self-review)#

This worker session has no sub-agent tool, so the critic brief (review-pack → CRITIC.md) was answered by the same session. It is labelled as a self-review, and a human second look is asked for before calling it shipped.

• Hook: at 1.5 s, a full-bleed Kirkjufell and "Ring Road" say "Iceland travel" at once. Frame 0 is the poster.
• Clarity and story logic: title (hook) → map (the whole loop) → eight stops, each with a name, a number and the inset lit at its place → the loop closes on the map → credits. Each shot has one job. The whip run is the only speed change, placed where the stops are closest together.
• Readability: the 2-bar stops hold their label about 3.9 s. The 1-bar stops show a name (plus "Reynisdrangar" at Vík) for about 1.8 s, and check's hold test passes. Contrast passes check. Tall type is 43 px and up, inside the Reels-safe box.
• Craft: round 1 used one primary transition (light-leak), one accent family (whips, three in a row), a dissolve outro. Round 2 (below) cut the leak to two half-strength uses and made the others short crossfades. Round-1 findings fixed: the overlays hidden in windows (Iteration 4) and the scrim pop (Iteration 5). In 9:16 the inset card covers the upper-left corner of each photo (by design: the top band); round 2 re-framed Vík so the card no longer hides the stack.
• Honesty: no direction word, distance, drive time or "our trip". The line is schematic and never labelled as the road. Every caption is on the audit's allowed list. Route order checked against the coordinates: counterclockwise and complete (8 stops, closed back to Reykjavík).
• Remaining WARN: check's 17 px reading of the SVG map label (see QA).

Review round 2 (critic)#

A second critic pass read review packs of the shipped files (both aspects). Verdict: ship after fixes, no blockers; facts, licenses and credits clean. Every should-fix and the cheap polish items were applied in one re-render (final-4.mp4, ring-road-9x16-2.mp4), and each fix was checked with snap --compare at the cited time against the round-1 files:

check after the fixes: PASS on both aspects with the same single WARN as before (the SVG map label size). This critic round was run by a separate critic pass on the shipped files; a human second look is still welcome.

Review round 3 (critic)#

A third pass checked the round-2 fixes on both shipped files. All of them held except one regression, and the Vík move itself caused it. In 16:9 the card sat under the Vík label's bottom scrim (.label::before, 62 % of the frame height at 0.5-0.72 opacity). Both are .ov overlays at z-index: 10, and the labels come later in the page, so for about 1.8 s of the 2.27 s shot the card's land went from about 110/130/134 to 58/72/76 and the amber dot turned brown. Fix: .ov.inset { z-index: 11; }. After the re-render (final-5.mp4), the land at (1640, 800) is 108/128/133 at 23.2, 23.892 and 24.3 s. snap --compare against final-4.mp4 shows no other change at 9.175, 22.8 or 24.8 s. In 9:16 the card (top band) never meets the label scrim. The 9:16 project with the fix snapped against the shipped ring-road-9x16-2.mp4 at 13.709 and 23.892 s differs inside the card only by encode noise (mean 1-2/255, the same as the whole frame), so the 9:16 was not re-rendered. check after the fix: PASS with the same single WARN.

Sources and licenses#

Not shipped: the photo originals, the render work folders, the 153 MB master and the review packs (local only).

The project files on GitHub