@oggii_0 absurd premises are underrated for ai ads. nobody checks fingers when a spy is tasting snacks through binoculars, the joke buys you every small glitch
@tryatria_AI the reason huel's apology static lasts is that the apology is about a real complaint from their reviews. copy the hook without a real complaint behind it and it just reads like a trick
@simonecanciello how does it hold the face past 10 generations? the first few always match, the drift usually shows up once you change the outfit and the light at the same time
@aimikoda the blockout doing the motion is the part people will skip. same reason a locked frame of the empty set keeps a sitcom room from drifting between cuts. reference beats description every time
@DavidRunsAds made a sitcom ad for first day built on the same tension. it only holds if the skeptic gets the funniest lines. once the believer wins every exchange it starts to feel like an ad
1. generate the empty set first. one wide frame, no people. it goes in as @Image on every shot, so the couch and the counter never move between cuts.
2. lock the camera. ai drifts when the camera travels, and sitcoms barely move theirs anyway. the limitation looks like the genre.
3. write all the coverage in one prompt: [wide on all four] [medium on her] [single on the friend]. same generation, so the mugs stay put.
4. leave 1 second of silence after every punchline. laugh track goes there in the edit.
then ship it. a perfect thumb won't sell more cream.
this ad has a melting thumb at 1:08 and it still sold a ton.
nobody buying face cream is checking fingers. they're watching four friends on a couch.
here's how i'd build it:
1. grab a real still with the look you want. it's only there to be described, none of its pixels go in
2. have a vision model describe it as json: face, skin texture, hair, wardrobe, light, lens
3. edit the json to your brief and generate 5s, 9:16, 1080p, text only
4. scrub the take and pick the frame where the face is neutral and fully lit
5. that frame goes in as @Image on every shot, and the prompt only describes the action
for a two person scene, do it once per character with the same room description in both, so the light matches when you cut
i ran this clip through a forensic ai check before posting.
verdict: filmed. real actresses, real mic, 9.5/10 realism.
it's ai.
what sells it is the faces holding up across every cut. that comes from one step most people skip: write the frame as json, run a 5 second 1080p sample in seedance with text only, then use the best frame as the character reference for every shot.
how i run it below:
who's the least likely person to compliment your customer?
put the compliment in their mouth.
a creator saying "you look younger" is an ad.
someone whose job is to doubt her saying it is a plot.
how to write one below:
The $200 Claude Ultra plan gets you $15,000 of Opus 5.5 in a month, but only around $3,000 of Fable 5.5.
Even ignoring that Fable is restricted to 50% of your weekly limit, it burns faster per $ than Opus does.
More info below
@geentibb@qwertyu_alex it's probable because I signed a nda of the brands. I can't just post stuff that I create freely or send to you the original assets
@DerekColling1@qwertyu_alex if you audience is 15y old half brain teenagers, sure, the pacing is terrible. but the audience isnt like that. you have to edit with intention, not guessing stuff