Wonder v38 — the in-image CTA button is gone

1 · The CTA finding, now tested on a wider set

Obvi: 0 of 70 stored ads draw a shopping button; the offer lives in ad_creative_link_title, the headline field Meta itself renders under the image. Original write-up.

Wider test (GetHookd, 45 brands, 66 US static image ads, every one inspected as pixels): 6 ads (9.1%) do draw a button. 25 carry a non-clicking label pill, 35 carry nothing. Four more carry a bare drawn click instruction with no enclosing shape — so 10 of 66 have a drawn CTA of some kind.

Verdict: QUALIFY, not confirm. Dropping the button as the house default is well supported — nine ads in ten don't have one. Saying the category never does it is not supported. And the button-drawers are not only junk: the two highest-performing of the six are established DTC brands (perf 100 and 84), and both pills carry an offer or a category — BUY 1 GET 1 FREE, SHOP KINKY STRAIGHT — never a bare "Shop now". The two conventional "Learn More" pills both score 1/Testing. If we ever put a pill back, that is the shape worth testing.

Also learned: auditing only the bottom 38% of a frame undercounts by a third — two of the six buttons sat mid-frame. Full audit.

2 · What replaced the button

Obvi's own bottom furniture, rotated so no two neighbouring ads carry the same device: a proof line (6 ads), a thumbs guarantee (2), a spec row (2). It is type directly on the photograph — no panel, band, strip or scrim behind any of it.

Placement differs by ratio deliberately. The 4:5 runs in feed with nothing over the image, so the furniture is bottom-pinned exactly as Obvi runs it. The 9:16 loses its bottom 35% to the Reels caption and CTA overlay, so the same line sits under the support line where it is actually seen. Removing the button is what made that possible — it was the heaviest object in the frame and it owned the bottom third.

One open decision. Six of the ten carry no proof roundel. On W01/W05 the bottles run the full width of the frame and on W07/W08 two portraits run edge to edge, so every 188–212px roundel position was measured to land on the product label, a face or the before/after evidence. Filling the frame edge to edge and stamping a roundel on it are in direct conflict. Those six say the review count in the furniture line instead. If you want the roundel back on all ten, the plates have to give up some scene.

3 · The 4:5 crop windows

Every plate was measured against a pixel ruler. A 4:5 window is 1440 of the plate's 2048 rows, so any subject spanning more than 1440 rows cannot be cropped without cutting something. Three of the ten original plates failed that test — W03 (hair crown 60 → bottle base 1845, span 1785), W05 (cap 195 → reflection 1800, span 1605) and W09 (fingers 230 → bottle base 1890, span 1660). Any window either sliced her forehead or cut the bottle base.

They were reshot rather than cropped badly. The brief that worked states three pixel bands, not percentages: top band backdrop only, a core band holding the whole subject with slack, bottom band floor only — plus "shoot much wider than instinct". The first attempt still overshot by ~60px, so the core band was tightened to 420–1620 to leave room for the model's overshoot. The two before/after plates were reshot twice more: their heads sat mid-frame, so a hook either covered the parting — the actual evidence — or landed in the Reels reserve. They now sit in the bottom half with the upper half a lit, shadowed wall.

4 · Hooks — all ten lead with the pain

Rebuilt from a pain map, not from features, and each ad enters the tension from a different door: direct recognition, feared future, social exposure, a stalled plateau, a recurring cost, a hidden cause, a failed alternative, category negation, a physical complaint, a life-stage shock. Each card below names its pain and its entrance.

The claim boundary is unchanged and deliberate. The only bridge used is the qualified sentence already live on the PDP — batana oil may help with shedding, and conditions dry, fragile lengths. No timeframe, no regrowth promise, no "in 2 weeks". That is a conscious difference from Obvi, whose own H04/H06/H10 run "REGROW Hair in Just 2 Weeks".

No price, no discount, no offer anywhere in the artwork — that is the finding applied. The offer belongs in Meta's headline field.

5 · Gates — all clean, and what the gates do not catch

Two honest notes. First, the product gate measures the label, not the bottle — a generous bottle-shaped box made it fire on everything and meant nothing. Second, the gate only checks pills, badges and buttons, so it never looks at a plain headline: five readability defects passed every gate and were caught by eye (navy type crossing dark dropper caps on W01 and W10, navy over dark hair on W09, a navy spec row on a black top in W08-45, a white brand line on light pink in W01-45). All five were fixed by per-element ink, then re-measured.

Weakest ad: W10. Its headline still crosses the bottle's cap and its furniture sits on busy marigolds. Legible, gate-clean, but the least resolved of the ten — say the word and it gets the same three-band reshoot the others had.

What is real

The product is the actual bottle in every scene, generated with source/real-bottle-reference.png attached. The only numeric proof is the genuine 4.8/5 from 1,121 Trustpilot reviews, said once per ad. Free shipping and skip-or-cancel are live PDP terms. The spec row is true.

The two before/after plates are generated stand-ins and are labelled on the artwork — "ILLUSTRATION · NOT A CUSTOMER RESULT". That label is a compliance requirement, not decoration. No invented customer count, testimonial, timeframe or measured result anywhere.

Method: gpt-image-2.5-flare at quality=medium via /v1/images/edits. No text generated into artwork — all typography composited deterministically in Remotion.