Liverty shock batch v2 - rebuilt to the reference grammar
After your note that the overlay was far too large and far too frequent, the 22 reference videos were measured frame by frame and the three cuts were rebuilt to those numbers, and they play with sound once unmuted.
What the references do, measured. Across the 22 videos you listed, 1,130 seconds read at
one frame per second, a picture inside the talking frame appears 13 times for 20 seconds in total, which is under
2% of the runtime. The median one is 0.40 of the frame wide, none is taller than 0.28, and none stays longer than 3
seconds. Showing a thing is done by cutting to it full-frame for one to three seconds, or by holding it in the hand.
The podcast, telehealth, stage, street and multi-camera videos have no pictures at all. Our v1 cuts had 22 overlays
on 110 seconds, 91% of the clock covered, each one half the frame wide and held for ten seconds. v2 is at 12-15%
coverage, each picture small and gone within two seconds. The numbers live in
swipes/formats/specs/*.json, and the build now refuses a plan that breaks them.
s01 - the steroid treadmill
| on screen | 7 overlays, 12% of 109.8 s: the hook label, three two-second product cuts, one condition cut, one small skin picture, the clear-week label; the 92% line and the 90-night money back guarantee stay in the spoken script and the captions |
| product | full-frame for two seconds when it is first named, again on the dropper line, again at the close |
s02 - four creams a dermatologist will not use
| on screen | 9 overlays, 15% of 109.8 s: the hook label, the four list-number labels, two small pictures over the shoulder, one condition cut, one product cut |
| grammar | built to the comment-farming spec, the doctor-explains format, which allows a few more labels |
This one still needs your call on the ending. The script stops mid-sentence on
"so your barrier can settle and rebuild freely," with no close and no call to action. The donor transcript ends in
the same place, so there is no donor sentence available to carry an ending, and nothing has been written to fill
it.
s05 - what is inside your cream
| on screen | 8 overlays, 15% of 98.8 s: the hook label, the banned-in-Europe label, five two-second demo cuts (the two tubes, the water, the pan, the arm, the product), one small ointment picture |
The QA verdicts, and why two fatals stand
The full verdicts are in s01, s02 and s05.
Two findings are graded fatal and are left as they are, because they are facts about the product rather than
defects in the cut, and they are yours to rule on:
| the label | The reviewer reads the bottle's own label, "supports healthy liver function", against an eczema pitch and calls the premise broken. That is the real Phoilex label on the real product photo. If the drops are to be sold for skin, the label or the packshot is a decision about the product itself. |
| the close of s01 | The last two seconds show the store's own packshot with the box beside the bottle. The reviewer calls the box unapproved because the reference it was given is the bottle alone. Same product, the store's photo. |
| the competitor creams | The reviewer was written for product-swap remakes, where any competitor frame is a failure. Here the script's subject is the other creams, so the tubes, the jar and the pan are the point of those lines and stay. |
| the voice | The brand still comes out as "Liberty" or "every" in the audio on all three. The captions carry the right spelling. The fix is a different belt voice or a Redub on that one word. |
25 flagged claims across the batch are still waiting on you for a keep or cut decision.
Every one of them is still in the copy, as it should be until you say otherwise.
How the next ad in this format gets made
The structure of each format was already written up in research/x-frankyecom/RECIPES.md. What was
missing was the screen layer with numbers and a build that holds to it, and that is now
swipes/formats/SCREEN-GRAMMAR.md with one spec per format and two tools. A new talking-head ad is a new
locked script, a voice take, a lip-sync, captions on the cut box, a cue list of what to show, and the two tools;
nothing is re-derived.