Every Katechon campaign on the Voana account, in one place. Per campaign: full economics — AI generation cost beside ad spend beside revenue — every creative playable, lifetime Meta performance scored against benchmark, and generation provenance under “Under the hood”.
AI generation is the engine's render ledger (video, image, voice and music model calls). Excludes the KAT001 poster round, produced outside the ledgered pipeline.
Hook rates rose generation over generation. First-generation creatives measured 45–55%; every creative shipped after the swipe-library analysis measures 60–76%.
Narration corrected retention. Silent KAT005 held 3.4% of viewers; its narrated child holds 23.5% and converted at 2.18 ROAS.
Iteration cost is collapsing. A replacement hook on a proven body renders for ~$8 same-day, versus $44 for the original body.
Static poster · launched 2026-07-22

| Variant | Spend | Hook | Hold | CTR | CPM | CPC | Landing views | Checkouts | Purchases | ROAS |
|---|---|---|---|---|---|---|---|---|---|---|
| A | $21.22 | — | — | 6.1% ✓ | $108.0 | — | 13 | 7 | 0 | — |
| B | $33.44 | — | — | 6.3% ✓ | $105.0 | — | 18 | 5 | 0 | — |
Lifetime numbers from Meta · pulled 2026-07-29 11:06 ET · ✓ meets benchmark, ✓✓ strong.
Lowest-cost test in the trial: a single static frame. It recorded the best front-end economics of the first generation — 6.1–6.3% outbound CTR at $1.63–2.06 per landing-page view — and produced 12 initiated checkouts on $54.66 of spend with zero purchases. High CTR followed by checkout stalls indicates a funnel-side issue rather than a creative one. Rotated off after variant B crossed $50 without a sale.
Next Poster rounds remain the cheapest instrument for scouting desire angles (odor, intimacy, garment comfort) ahead of video spend.
Hook-poster round output (iteration #150). The poster engine now lives in a separate repository; source asset backfill pending.
73s marionette VSL · launched 2026-07-22
| Variant | Spend | Hook | Hold | CTR | CPM | CPC | Landing views | Checkouts | Purchases | ROAS |
|---|---|---|---|---|---|---|---|---|---|---|
| A | $8.42 | 45.7% | — | 0.9% | $77.0 ✓ | — | 1 | 0 | 0 | — |
| B | $19.07 | 49.3% | — | 4.6% ✓ | $109.0 | — | 7 | 0 | 0 | — |
Lifetime numbers from Meta · pulled 2026-07-29 11:06 ET · ✓ meets benchmark, ✓✓ strong.
Retention — of viewers who start, how many are still watching:
Art-style test: a 73-second marionette VSL probing whether a non-clay world could match the clay system. Hooks measured 45.7% / 49.3%, under the 50% benchmark. Variant A under-delivered (0.92% CTR on $8.42); variant B's 4.57% CTR had no downstream support. Rotated off under the verdict floor.
Next Marionette spine retired. Novelty budget is redirected to new hooks and desire angles inside the clay system.
Spine transplanted from the swipe library — hookd board ad in the marionette/creature family. 13 scenes, VO body.
Story structure — 13 engine scenes (★ = hero shot, green = pro-tier render):
From the spec — the hook scene's still prompt (excerpt):
${SPINE}. On ${STAGE}: close on ${MAE_DOLL} looking down at her wooden elbow joint, which glows an angry sore red with three little cartoon squeak-lines radiating from it; her painted face in a big sad wince; nothing else in frame; the red joint is the single unmistakable subject.…
73s clay VSL · launched 2026-07-22
| Variant | Spend | Hook | Hold | CTR | CPM | CPC | Landing views | Checkouts | Purchases | ROAS |
|---|---|---|---|---|---|---|---|---|---|---|
| A | $10.98 | 54.7% ✓ | — | 5.7% ✓ | $157.0 | — | 6 | 3 | 0 | — |
| B | $9.72 | 55.4% ✓ | — | 7.1% ✓ | $139.0 | — | 2 | 1 | 0 | — |
Lifetime numbers from Meta · pulled 2026-07-29 11:06 ET · ✓ meets benchmark, ✓✓ strong.
Retention — of viewers who start, how many are still watching:
A structurally strong body shipped with a known-weak hook: the internal ship gate scored the hook take 3 against a floor of 8, and it shipped on a recorded waiver to get the body into market. Market data matched the gate: 54.7% / 55.4% hooks against 5.7% / 7.1% CTR and 4 checkouts on only $20.70 of spend. Spend was halted early and redirected into surgery — two replacement hooks spliced onto the bit-exact body at $7.45 and $8.05 per child, versus $44.17 for the original render.
Next KAT007/008 run the controlled hook experiment; the winner inherits the body.
Second-generation clay VSL. Its hook shipped on a gate waiver and was later replaced twice (KAT007/008) on the same body.
Story structure — 25 engine scenes (★ = hero shot, green = pro-tier render):
From the spec — the hook scene's still prompt (excerpt):
${SPINE}. Morning light through a frosted window in a cozy miniature 1970s clay bathroom — avocado-green tiles, blush-pink sink, brass fixtures: ${MAE} stands at the mirror mid-wince, fanning her collar away from her chest with a small woven fan, cheeks flushed; beside her the mirrored medicine cabinet hangs open, crammed to bursting with dozens of little BLANK unlabeled clay jars, tubes and pump bottles teetering on the shelves, one white tube frozen mid-…
53s educational VSL · launched 2026-07-22
| Variant | Spend | Hook | Hold | CTR | CPM | CPC | Landing views | Checkouts | Purchases | ROAS |
|---|---|---|---|---|---|---|---|---|---|---|
| A | $35.87 | 48.2% | 14.2% | 3.9% ✓ | $106.0 | $2.39 | 10 | 6 | 1 ($27.41) | 0.76 |
| B | $39.99 | 55.4% ✓ | 16.8% | 5.1% ✓ | $102.0 | $1.6 | 13 | 8 | 0 | — |
Lifetime numbers from Meta · pulled 2026-07-29 12:24 ET · ✓ meets benchmark, ✓✓ strong.
Retention — of viewers who start, how many are still watching:
The only first-generation ad to reach the $30–50 verdict floor on both variants. A converted ($27.41; lifetime ROAS 0.76 at $10/day budgets). B did not convert but recorded the trial's strongest intent signal: 8 initiated checkouts from 13 landing views at 5.1% CTR. Consistent with the account-wide pattern, the lower-CTR variant is the one that converted. Hooks 48.2% / 55.4%; hold in the mid-teens.
Next The timeline-proof structure (mechanism → six-week timeline) qualifies for a hook-swap round before further body spend.
Spine from hookd 90372054 — at 305 days active, the longest-running ad in the 468-ad swipe library.
Swipe provenance: hookd-176699/90372054
Story structure — 21 engine scenes (★ = hero shot, green = pro-tier render):
From the spec — the hook scene's still prompt (excerpt):
${SPINE}. Tight handheld close-up in a bright morning bathroom: ${JUNE} winces as she presses the back of her hand against the side of her ribcage through a thin cotton shirt, mid-flinch, eyes down, jaw tight — the private, instantly-recognizable discomfort gesture; behind her a mirror cabinet crammed with BLANK unlabeled jars and tubes; raw, candid, scroll-stopping intimacy.…
49s silent escalation montage · launched 2026-07-22
| Variant | Spend | Hook | Hold | CTR | CPM | CPC | Landing views | Checkouts | Purchases | ROAS |
|---|---|---|---|---|---|---|---|---|---|---|
| A | $64.32 | 61.8% ✓ | — | 3.8% ✓ | $62.0 ✓ | — | 25 | 4 | 1 ($40.91) | 0.64 |
| B | $60.08 | 62.2% ✓ | 3.4% | 3.5% ✓ | $48.0 ✓ | $1.0 | 30 | 8 | 0 | — |
| C-7R | $18.73 | 62.9% ✓ | 23.5% | 3.5% ✓ | $65.0 ✓ | $1.44 | 10 | 2 | 1 ($40.91) | 2.18 |
| C-Comp small sample | $5.46 | 56.9% ✓ | 15.9% | 1.1% | $62.0 ✓ | $1.82 | 1 | 0 | 0 | — |
Lifetime numbers from Meta · pulled 2026-08-02 17:53 ET · ✓ meets benchmark, ✓✓ strong.
Retention — of viewers who start, how many are still watching:
First conversion of the trial, two days after launch at $14.48 of spend. Hooks 61.8% / 62.2% with the trial's best CPMs ($48–62). The retention curve exposed the failure mode: hold 3.4%, with viewers exiting immediately after the hook. Diagnosis — post-hook drop-off, early product reveal, no narration. Both corrections shipped within 72 hours: KAT006 re-authors the body; variant C overlays direct-response narration and captions on the corrected body.
C is converting: on the 7R funnel, a $40.91 sale on $18.73 of spend — ROAS 2.18 — with hold up from 3.4% to 23.5%. The chart compares the silent original against C-7R.
Next C-7R is the scale candidate. Silent v1 retired.
Spine from hookd 53535539 (260 days active, 5 concurrent placements). Silent by design: BEATS is empty — no VO, no captions, the music carries a 6-chapter arc.
Swipe provenance: hookd-176699/53535539
Story structure — 14 engine scenes (★ = hero shot, green = pro-tier render):
From the spec — the hook scene's still prompt (excerpt):
${SPINE}. Extreme handheld macro, stark white bathroom background: the crease of an inner arm where skin rests against skin — ${SKIN} — flushed an angry irritated red with fine chafed texture, raw, real, uncomfortable to look at but intact skin, no broken or weeping surface; caught mid-flinch, the arm tensed; no product, no brand, no text; pure unexplained pathology as the scroll-stopper.…
47s silent montage, story lead · launched 2026-07-29
| Variant | Spend | Hook | Hold | CTR | CPM | CPC | Landing views | Checkouts | Purchases | ROAS |
|---|---|---|---|---|---|---|---|---|---|---|
| 7R small sample | $3.04 | 75.8% ✓✓ | 18.1% | 2.8% | $42.0 ✓ | $0.76 | 2 | 1 | 0 | — |
| Comp small sample | $3.44 | 66.7% ✓✓ | 12.8% | 7.7% ✓ | $88.0 ✓ | $1.15 | 3 | 0 | 0 | — |
Lifetime numbers from Meta · pulled 2026-08-02 17:53 ET · ✓ meets benchmark, ✓✓ strong.
Retention — of viewers who start, how many are still watching:
The KAT005 correction applied at the body level: hook frozen (market-validated, untouched), the 3–18s lead re-authored as a narrative sequence — false hope → betrayal → verdict — and the first brand moment moved to 18.2s per the funnel-stage rule. Early signal is strong for $6.48 of spend: 75.8% hook on 7R, the highest in the trial, at a $42 CPM, the lowest in the trial. Sample is small (72 + 39 impressions); no verdict before the floor.
Next Highest-priority spend allocation in the trial.
SPEC_VERSION 2 of the rawmacro spec. Hook scene cached from v1 and locked against edits; every lead shot ≤3s with one readable action.
Swipe provenance: hookd-176699/53535539
Story structure — 14 engine scenes (★ = hero shot, green = pro-tier render):
From the spec — the hook scene's still prompt (excerpt):
${SPINE}. Extreme handheld macro, stark white bathroom background: the crease of an inner arm where skin rests against skin — ${SKIN} — flushed an angry irritated red with fine chafed texture, raw, real, uncomfortable to look at but intact skin, no broken or weeping surface; caught mid-flinch, the arm tensed; no product, no brand, no text; pure unexplained pathology as the scroll-stopper.…
73s clay VSL — hook swap · launched 2026-07-29
| Variant | Spend | Hook | Hold | CTR | CPM | CPC | Landing views | Checkouts | Purchases | ROAS |
|---|---|---|---|---|---|---|---|---|---|---|
| 7R small sample | $6.49 | 60.0% ✓ | 22.7% | — | $148.0 | — | — | 0 | 0 | — |
| Comp small sample | $6.61 | 64.1% ✓ | 32.7% | 10.2% ✓ | $135.0 | $0.94 | 2 | 0 | 0 | — |
Lifetime numbers from Meta · pulled 2026-08-02 17:53 ET · ✓ meets benchmark, ✓✓ strong.
Retention — of viewers who start, how many are still watching:
Hook-swap child #1: “It's not your skin. It's your cream.” spliced onto the KAT003 body; total render cost $7.45 because only the hook is new. Early reads: hooks 60.0% / 64.1%, holds 22.7% / 32.7%, and 10.2% CTR on the Comp funnel from 49 impressions (directional only). QA note on record: the hook's micro-motion is barely perceptible; if the hook underperforms at spend, the correction is stronger motion, not additional takes.
Next Run to the floor against KAT008; the winner inherits the body.
Child of KAT003 via /redo: hook HPAT-005 from the swipe-library hook bank, spliced onto the cached body. Bit-exact body means the A/B is purely the hook.
Swipe provenance: HPATS.md HPAT-005 Misdiagnosis Reversal (Σ852d, claymation-proven)
Story structure — 25 engine scenes (★ = hero shot, green = pro-tier render):
From the spec — the hook scene's still prompt (excerpt):
${SPINE}. A single BLANK glossy white cream jar under a harsh noir interrogation spotlight on the blush-pink clay bathroom counter, the cozy 1970s pastel room falling away into dramatic darkness at the edges: ${MAE} looms over the jar leaning in nose-to-lid like a detective mid-accusation, one chunky clay finger pointed hard at it, eyes narrowed in huge theatrical suspicion, eyebrows a dramatic V; the jar's lid sits ajar with a single glossy bead of cream …
73s clay VSL — hook swap · launched 2026-07-29
| Variant | Spend | Hook | Hold | CTR | CPM | CPC | Landing views | Checkouts | Purchases | ROAS |
|---|---|---|---|---|---|---|---|---|---|---|
| 7R | $42.45 | 65.5% ✓✓ | 20.5% | 4.5% ✓ | $128.0 | $2.23 | 16 | 3 | 1 ($35.96) | 0.85 |
Lifetime numbers from Meta · pulled 2026-08-02 17:53 ET · ✓ meets benchmark, ✓✓ strong.
Retention — of viewers who start, how many are still watching:
Hook-swap child #2: “Stop putting cream on that rash.” First of the pair to reach the verdict floor ($42.45) and converted: a $35.96 purchase at ROAS 0.85 lifetime, with a 65.5% hook — above the 65% strong threshold — 4.5% CTR and 20.5% hold. The Comp copy received no delivery; Meta allocated the ad set's budget to 7R. The direct-command hook currently leads the inheritance experiment.
Next Continue past $50 monitoring CPA; scale-set candidate alongside KAT005-C.
Child of KAT003 via /redo: hook HPAT-006. Same cached body as KAT007 — the pair is a controlled hook experiment.
Swipe provenance: HPATS.md HPAT-006 Incumbent Remedy Indictment (Σ705d)
Story structure — 25 engine scenes (★ = hero shot, green = pro-tier render):
From the spec — the hook scene's still prompt (excerpt):
${SPINE}. Tight dramatic two-shot at the blush-pink clay bathroom vanity, flat morning light: ${MAE} frozen mid-application — her chunky clay fingers loaded with a glossy dollop of moisturizer from a BLANK open white jar, hovering an inch above her other forearm — while from off-frame the teal-scrubbed arm of ${NURSE} reaches in and catches Mae's wrist in a firm gentle grip; Mae's eyes gone huge with silent-film shock, eyebrows sky-high, mouth a small O; t…
54s narrated story film · rendered 2026-07-28, not yet launched
The first creative with a designed narrator: a 54-second single-protagonist story film — ritual, betrayal, verdict, product, payback. Two full review cycles are incorporated: re-scored music, a restored narrative beat, the product reveal aligned to the narration line it lands on, and a sterile product shot replaced with a naturalistic one. Premium render — $75 ledger on a per-scene veo-3.1 + kling model mix.
Next Rendered and delivered; not yet launched. Proposed slot: 7R funnel at $10/day inside a hook × scroll-stopper set.
Original script (“drydown”), not a swipe transplant. Voice: designed Elaine (ElevenLabs). Models mixed per scene after QA: veo-3.1 for portraits/textures, kling 2.5 for directed action.
Story structure — 19 engine scenes (★ = hero shot, green = pro-tier render):
From the spec — the hook scene's still prompt (excerpt):
${SPINE}. Extreme handheld macro, stark white bathroom tile, harsh honest morning light: ${CREASE} exactly as in the TOP panel of the attached arm reference — the same crease, same broad diffuse redness, same finely lined hand — with the round nozzle of a plain white blow dryer entering from the frame edge aimed point-blank into the crease, and two fingertips of the matching mid-sixties hand already descending toward the red skin; no product, no brand, no …
Voice: Designed narrator (ElevenLabs custom voice, built for this film)
Music: Six-chapter underscore, dread-to-warmth arc
Models: Mixed per scene after QA: veo-3.1 for portraits, textures and matte glides; kling 2.5 turbo pro for directed-action scenes
Hook design: Blow-dryer-at-crease wrongness + press-dent flinch + red headline card “Skin Fold Rash Won't Go Away?”
AI generation is the engine's render ledger (video, image, voice and music model calls), including the 2026-08-04 client fix round. No ad spend yet — the campaign has not launched.
Nine films in one overnight flight, every run under its cap. Three sections — claymation, realistic, moonshot — rendered in parallel by a nine-agent fleet. Generation ledger $116.38 total, $12.93 average per film against a $16 hard cap and a $20 authorization; campaign #1's single most expensive body was $44.
A client review round now costs dollars, not re-renders. All nine films were reviewed frame-by-frame; seven took fixes — new hooks, label re-shoots, script patches — as surgical retakes on cached runs. The entire fix round cost roughly $8 of incremental generation and shipped same-day.
The label doctrine flipped. Client direction: the real Voana label, legible, in every prominent product shot. Where generation mangled the label's micro-type, real product pixels were composited in — every shipped wordmark is letter-perfect on frame-level inspection.
48s claymation VSL — base script · rendered 2026-08-04, not yet launched
The claymation base-script slot: CORE-1 narrated verbatim over the clay system that produced campaign #1's purchases. Client review rejected the original hook image and a wooden prop in the reveal; the fix round replaced the hook outright (a glossy cream ribbon frozen mid-squeeze above blushed clay skin) and re-shot the reveal with the real labeled product as a photographic insert. Verified frame-by-frame after the retake: label letter-perfect, story coherent end to end.
Next Launch candidate for the clay lane; the hook is unproven in market and should run inside a hook-test set.
Clay base-script slot: KAT003/KAT008 clay-VSL system × HPAT-006 direct-command hook × KAT002's seal imagery transplanted to clay. Reveal + endcard are real-pixel product photography inside the claymation.
Swipe provenance: KAT003/KAT008 clay-VSL system × HPAT-006 direct-command hook × KAT002 marionette seal imagery
Story structure — 14 engine scenes (★ = hero shot, green = pro-tier render):
40s claymation comedy mini-episode · rendered 2026-08-04, not yet launched
Entertainment-first mini-episode: villain mascots — the sweat drop, the cream goo, the stink cloud — party in a clay canyon until one matte powder road evicts them. Rendered with zero concessions on the first pass and approved unchanged in client review — the only film of the nine to take no notes.
Next Strongest pre-launch signal in the claymation section; propose it leads the lane at launch.
Spine from hookd 111594939/111594935/111595003 — the claymation-story family (104–112-day survivors) × HPAT-002 gross-out hook. Villain-mascot world built for entertainment-first watch time.
Swipe provenance: hookd 111594939/111594935/111595003 (claymation-story family) × HPAT-002 clay gross-out hook
Story structure — 13 engine scenes (★ = hero shot, green = pro-tier render):
45s claymation mockumentary · rendered 2026-08-04, not yet launched
Deadpan police-lineup mockumentary: four incumbent products confess why they fail, a detective delivers the verdict, and the stick descends into the spotlight. Two surgical retake rounds during generation removed a hallucinated wooden applicator from the reveal; client approved the delivered cut.
Next Runs the incumbent-indictment desire angle against KAT011's pure entertainment in the clay lane.
Spine from hookd 102373716/102373756 (HPAT-012 incumbent self-own) + 86556057 (villain confession). The failed-alternatives block is the engine of the cgi-mascot family (n=185).
Swipe provenance: hookd 102373716/102373756 (HPAT-012 incumbent self-own) + 86556057 (HPAT-010 villain confession)
Story structure — 15 engine scenes (★ = hero shot, green = pro-tier render):
46s realistic macro VSL — base script · rendered 2026-08-04, not yet launched
The scale slot: the fleet's proven raw-macro attention system carrying the proven script at campaign-2 craft. Client review flagged the unlabeled stick at the brand moment; the fix round re-spun reveal, glide and proof shots with the real label legible exactly as the brand line lands — micro-type transplanted from real product pixels where generation garbled it.
Next Direct heir to the KAT005/006 system that produced hooks in the 60–76% band; first launch priority in the realistic lane.
The KAT005/006 raw-macro attention system × hookd 53535539 (260 days active). Label posture flipped in the fix round: authentic Voana label only, micro-type as real ink pixels.
Swipe provenance: KAT005/KAT006 rawmacro system × hookd 53535539 (260d before-after-demo spine)
Story structure — 17 engine scenes (★ = hero shot, green = pro-tier render):
48s confessional testimonial · rendered 2026-08-04, not yet launched
First video on the restored-intimacy desire axis — the bra hangs on the closet door for two summers and the proof is its absence. Fix round: both product moments re-shot label-facing and the CTA line sharpened to “If your skin rash keeps coming back…” so the sentence carries its own referent.
Next Tests the intimacy axis untouched by campaign #1 video; watch hold rate on the VO-silent gap scene.
Spine from hookd 84007809 (241-day confessional) × HPAT-014 vulnerable-confession cold open. Prop law: the bra is always an object, never worn — proof is its absence.
Swipe provenance: hookd 84007809 (ugc-testimonial confessional, 241d) × HPAT-014 vulnerable-confession cold open
Story structure — 15 engine scenes (★ = hero shot, green = pro-tier render):
43s match-cut timeline VSL · rendered 2026-08-04, not yet launched
A calendar-sticker ritual compresses the six-week recovery claim into thirty seconds of match cuts. Client review caught a visual hallucination at the seven-second mark — retaken — plus the skin-rash CTA patch. Internal ship gate had already passed it 15/15 on hard checks.
Next Pairs with KAT004's floor-reaching timeline structure; candidate for the qualified-hook slot in the realistic lane.
Spine from hookd 90372059/90372054 (259/305-day countdown VSLs) × HPAT-001 qualified hook × our own KAT004 six-weeks system. Survived the seedance duration-floor incident with every cached take preserved.
Swipe provenance: hookd 90372059/90372054 (day-countdown milestone VSL, 259/305d) × HPAT-001 qualified hook × KAT004 six-weeks system
Story structure — 17 engine scenes (★ = hero shot, green = pro-tier render):
49s mock-documentary — moonshot · rendered 2026-08-04, not yet launched
Moonshot lane 1: an evidence-board documentary about a rash that returns every day at 2 P.M. Fix round re-shot the exhibit product shot with the real label squarely facing camera and split the new skin-rash CTA across caption cards that fit the frame.
Next The strangest structure in the campaign — a pure scroll-stopper test. Verdict quickly at low spend.
Moonshot lane 1 — mock-documentary case-file form. Evidence-board world; the daily-2-P.M. pattern gives the mechanism claim a narrative clock.
Swipe provenance: mock-documentary/case-file family — moonshot lane 1
Story structure — 16 engine scenes (★ = hero shot, green = pro-tier render):
49s spoken-verse ballad — moonshot · rendered 2026-08-04, not yet launched
Moonshot lane 2: a talking-blues ballad over a warm Americana bed, third-person testimonial told in verse. Client called it a great ad with one note — the blank stick — and the fix round swapped the real labeled product into both flagged scenes without touching composition or pacing.
Next Novelty spend candidate; if the hook holds above 60% the ballad family earns a second instantiation.
Moonshot lane 2 — ballad family (hookd 105370525, 113752119/113752155) instantiated for older-female skincare. Spoken-verse fallback after the music stage couldn't time sung lyrics.
Swipe provenance: hookd 105370525 (Tradesman Shred) + 113752119/113752155 (Quasi) — ballad-to-skincare instantiation; moonshot lane 2
Story structure — 14 engine scenes (★ = hero shot, green = pro-tier render):
27s macro-ASMR texture film — moonshot · rendered 2026-08-04, not yet launched
Moonshot lane 3: the dry-matte finish argued entirely through texture and foley — glossy rival cream against the stick's powder line, six whispered lines total. Client loved it; the fix round put the real label on every generated product appearance. The cheapest film of the fleet at $9.53.
Next Shortest runtime in the campaign — a natural fit for the cheap-verdict slot and for cutdown experiments.
Moonshot lane 3 — zero-VO macro family (hookd 46829112, HPAT-009) × silent carousel × raw-macro attention economics. Two-material world; the only gloss allowed is the rival cream.
Swipe provenance: hookd 46829112 (zero-VO object-carry macro, HPAT-009) × 107246378 (silent carousel) × KAT005/006 attention economics; moonshot lane 3
Story structure — 7 engine scenes (★ = hero shot, green = pro-tier render):
The ad-studio generation ledger has not been settled into per-film costs yet — AI generation shows only the posted lip-sync line until it lands. No ad spend — the campaign has not launched.
One winning script, two narrators. Both films run the same corpus-winning UGC script: KAT019 first-person as Mae, KAT020 flipped to second person and fronted by the @nurse toy mascot — a controlled test of narrator framing on an identical persuasion spine.
The mascot now actually speaks. KAT020's corner nurse is no longer a canned loop: a lip-sync engine animates the toy figure to the cut's own dialogue track and returns a transparent-background file that composites without keying — mouth frame-accurate to the VO for the full 58 seconds.
Produced in the new studio pipeline. Both cuts came out of katechon-ad-studio (shot boards → still heroes → image-to-video), with word-synced burned captions and the claymation pedestal endcard carried over from campaign 1's converting system.
56s claymation UGC VSL — winning script, first person · rendered 2026-08-10, not yet launched
The base slot of the winning-script transplant: UGC-script-01 from the winners corpus narrated verbatim (one covenant deviation — the product callout corrected to the anti-friction stick), first person as Mae. Hook 2 (“I stopped wearing bras because of this”) cold-opens on the bra-rash reveal; 24 scenes cut to the VO lines with word-synced captions and the claymation pedestal endcard.
Next Launch inside a campaign-3 hook-test set; its A/B against KAT020 isolates narrator framing on an identical script spine.
First film out of the katechon-ad-studio pipeline (shot.yaml boards → nano-banana multi-ref heroes → seedance-1-pro i2v). Body verbatim from the winners corpus (UGC-script-01); the 14 @rash closeups tripped Seedance's image-level moderation and were rescued on wan-2.2. Voice: ElevenLabs Sarah.
Swipe provenance: winners corpus UGC-script-01 (katechon-ecom/corpus/winners/winning-scripts.txt)
Voice: ElevenLabs Sarah — first-person Mae confessional (eleven_multilingual_v2)
Models: nano-banana multi-ref heroes → seedance-1-pro i2v, 720p locked camera; @rash closeups rescued on wan-2.2-i2v-fast after Seedance image-level moderation (E005)
Hook design: Hook 2 — “I stopped wearing bras because of this”; cold-opens on the bra-rash reveal
59s claymation VSL — nurse narration, lip-synced mascot · rendered 2026-08-10, not yet launched
The same script flipped to second person and fronted by the @nurse toy mascot, riding the lower-right corner for the full runtime. The mascot is genuinely speaking: a lip-sync engine animates the toy figure to the cut's own dialogue track and returns a transparent-background layer that composites without keying — mouth frame-accurate to the VO, exiting when the narration ends. Hook 3 (“Sudocream, deodorant, Powder, just to wear a bra?”) fronts 25 scenes.
Next A/B against KAT019 to read whether an authority mascot narrator beats a first-person confessor on the same persuasion spine.
Same UGC-01 body flipped to second person for the @nurse toy mascot (V2 covenant deviation, 2026-08-10). Corner narrator: HeyGen Avatar IV lip-sync of the nurse talking pose to the cut's own dialogue track, transparent-WebM composite — no chroma key. Voice: ElevenLabs Bella.
Swipe provenance: winners corpus UGC-script-01 × cgi-mascot-vsl family (n=185, hookd swipe library)
Voice: ElevenLabs Bella — @nurse narration, second-person POV flip of the same script (eleven_multilingual_v2)
Models: nano-banana multi-ref heroes → seedance-1-pro i2v, 720p locked camera; corner narrator: HeyGen Avatar IV lip-sync of the @nurse talking pose to the cut's own dialogue track, transparent-WebM output composited without chroma key
Hook design: Hook 3 — “Sudocream, deodorant, Powder, just to wear a bra?”