Floom Film v16 — Visual-vs-VO Timing Audit

69.3s cut @ 30fps · frames pulled at each key VO word (per v10CaptionWords.ts) · every frame read visually

Verdict. The product section (L08–L12) and the close (L13–L16, LCTA) are tightly synced — every card, orb, and reveal lands on its word. The problems are concentrated in the crash→turn transition (L03–L04) and the Floom name reveal (L07).

Two P0s: (1) the manual-work spreadsheet plate arrives ~2.5–3s after the word “manual,” and (2) the word “Floom” is spoiled by the caption + corner mark ~3.3s before it’s spoken.

P0 breaks itP1 noticeableP2 minor

Timing / visual mismatches (prioritized)

1 P0 L03→L04 13.94s → 17.0s
Spoken: "manual" / "work" (13.94–14.54s)

What IS on screen

At the word manual/work the screen shows the failed OpenClaw card + empty photoreal room. The literal manual-work visual (the report_by_hand.xlsx spreadsheet being typed by hand) does NOT appear until ~16.4–17.0s, during the next line “we realized this couldn’t be it.”

What SHOULD be

The by-hand spreadsheet plate should land ON “manual work” (~13.9–14.5s), not ~2.5–3s later on L04. Right now the punchline visual for “manual” arrives after the word is gone.

Fix. Pull the report_by_hand.xlsx plate earlier so it reveals on “manual” (~13.9s), inside L03, before the L04 crossfade. Or move the “manual work” VO/caption to meet the plate.
13.94s ‘manual’ — no manual-work plate yet
13.94s ‘manual’ — no manual-work plate yet
16.4s ‘this’ — spreadsheet only now fading in
16.4s ‘this’ — spreadsheet only now fading in
17.0s ‘couldn’t’ — report_by_hand.xlsx fully in (late)
17.0s ‘couldn’t’ — report_by_hand.xlsx fully in (late)
2 P0 L07 21.9s (word “Floom” at 25.19s)
Spoken: "and" / "two" (21.4–22.6s)

What IS on screen

The caption already spells out “we’re launching Floom.” with “Floom” pre-highlighted in a blue box, AND the Floom brand mark sits in the top-left corner, from the very start of the beat (~21.9s) — ~3.3s before “Floom” is spoken (25.19s).

What SHOULD be

The name “Floom” should be revealed on the word “Floom” (25.19s). The big centre logo reveal IS correctly timed there — but the early caption + corner mark spoil it, so the reveal has no punch.

Fix. Hold the caption on “Two months later,” and reveal “we’re launching Floom.” word-by-word so “Floom” lights up at 25.19s. Delay/omit the corner Floom mark until the centre logo reveal fires.
21.9s — caption + corner Floom already shown
21.9s — caption + corner Floom already shown
22.55s ‘two’ — name still spoiled 2.6s early
22.55s ‘two’ — name still spoiled 2.6s early
3 P1 L04 15.4–17.7s
Spoken: "we realized this couldn’t be it"

What IS on screen

L04 is authored as a founder full-frame (mode:“full”, L04LowFull) but renders as a flat vector/cartoon illustration (flat chair, flat beige wall, flat green shrub) with the founder shrunk to a PIP. It clashes hard with the photoreal 3D rooms of L02/L03.

What SHOULD be

Either the founder reflective full-frame (per founderLines) OR a consistently photoreal plate. The flat-illustration look is an outlier in the film’s visual language.

Fix. Swap L04 background to the intended photoreal low/reflection plate (or the founder full-frame). If the illustrated plate is deliberate, restyle it to match the 3D room render.
15.74s ‘realized’ — flat cartoon plate, off-register with the film
15.74s ‘realized’ — flat cartoon plate, off-register with the film
4 P2 L14 56.1–59.5s
Spoken: "people spending time on work they shouldn’t have to do"

What IS on screen

The burned-in scene headline reads “stop doing work you shouldn’t.” — which is NOT the spoken line (“people spending time on work they shouldn’t have to do”). This beat also drops the word-by-word caption style used everywhere else.

What SHOULD be

Caption text should match the VO, or the beat should clearly be an editorial headline. Right now it reads as a caption that disagrees with the voiceover.

Fix. Either match the headline to the VO wording, or keep it as an intentional editorial title but make that stylistically distinct so it doesn’t read as a mis-transcribed caption.
56.71s ‘time’ — headline says ‘stop doing work you shouldn’t’ (≠ VO)
56.71s ‘time’ — headline says ‘stop doing work you shouldn’t’ (≠ VO)

Beats that ARE correctly timed

L010–2sOpenClaw lobster badge lands on “openclaw/scammed”; money emoji on “scammed us.”
L022–10sOrbs = the “vision” on “sold”; OpenClaw online chat on “24/7 background”; WhatsApp chip on “whatsapp”, Telegram chip added on “telegram”. Tight.
L0310–13sOrbs go dead + OpenClaw ‘disconnected’ land on “broke.” Error log builds through the beat. (Only the manual-work PAYOFF plate is late — see #1.)
L0517.8–20s“the vision was right / the infrastructure wasn’t” text card matches the L05 VO.
L0620–21.4sFounder full-frame on “so we started building.” Correct.
L07 centre25.19sThe BIG centre Floom logo reveal fires exactly on the word “Floom.” Correct (it’s the early caption/corner mark that spoils it, #2).
L0825.6–29.4sTwo orbs + ‘FLOOM ORCHESTRATOR’ on “platform.” Correct.
L0929.7–32.3sSpawn card + model chips + typed prompt land on “spawn” then “prompt.” Correct.
L1032.6–40.7sModel choreography is excellent: approval card on “approval” (35.57), schedule card on “schedules” (38.94), trigger cards on “triggers” (40.16). All on-word.
L1141–45.7s‘run log’ on “logged”; ‘Secured/encrypted’ on “secured”; ‘Sandboxed/isolated’ on “sandboxed.” All on-word.
L1246.4–53.5s‘Zero lock-in’ card on “lock-in”; model logos on “model agnostic”; GitHub ‘Open source’ card on “open source.” All on-word.
L1353.5–55.7sFounder full-frame on “it’s about one thing.” Correct.
L1559.4–60.8s‘This is Floom.’ brand card on-screen exactly during the “This is Floom” line. Correct.
L1660.8–66.5sGolden field payoff + ‘time is life’ headline land on “time is life” (61.19–62.37). The extended tail (post-VO) is intentional. Correct.
LCTA66.5–69.3sClean Floom mark + floom.dev closing card. Correct.