VidSpy

🎬 Briefs (2 pending — approve/reject gates everything downstream; verdicts sync on next run)

PENDING One Photo, Full Dance AI Video Generator · 114d · ↻2 · ✂1

Hook: A flat, obviously-static selfie suddenly breaks into a full-body dance on beat — the impossible motion in frame one is the scroll-stop, not the dancing itself.

Why it works: It sells identity-participation, not software: the viewer sees their own exclusion from a trend they can't perform dissolved by a single asset they already own, so the ad is a proof-of-capability demo rather than a claim, and the before-state (a photo) is universally possessed.

Adaptation: Build a 10s 'static-to-motion' proof reel where the first frame is unmistakably a still photograph — visible photo grain, frozen crowd, a hand mid-gesture — that erupts into rhythmic full-body motion on the downbeat, using only original casting/stills we generate or license. Cycle four visually distinct subjects (different ages, body types, settings) so the claim reads as universal rather than cherry-picked, then cut to a 2-second real screen capture of the Vid Gen AI upload→template→render flow so the mechanism is shown, not asserted. Use a generic high-energy original track and a generic movement style rather than a named choreography, and let the typography carry the trend framing so the creative stays evergreen when the underlying meme dies.

Risks: Likeness is the main exposure: animating a real, identifiable person's photo into a dance implies endorsement and, in several US states plus GDPR jurisdictions, needs explicit consent — use only originally generated or fully released subjects, never scraped social photos or anything resembling a public figure. Music is the second: the actual trending audio is almost certainly not licensable for paid ads, so use an original or licensed track and let typography reference the trend instead. Named dances and their taglines can carry trademark/choreography claims — don't name the dance, don't reproduce a signature routine step-for-step. Meta will also scrutinize AI-generated-people ads under its manipulated-media and personal-attribute rules, so avoid 'make anyone dance' phrasing that implies uploading photos of third parties; frame it as your own photo. Finally, the demo must match real product output — showing renders the app can't actually produce is both a policy problem and a refund problem.

Lineage tag: vgai-c1575998583963829 ← put this in the shipped ad's name/UTM

packet.json ↗

PENDING One Photo, Instant Dancer AI Video Generator · 114d · ↻9 · ✂4

Hook: A dead-still, ordinary photo — a grandma, a pet, a coworker's LinkedIn headshot — suddenly breaks into a full-body dance on beat, with zero cut between still and motion so the brain registers 'that photo just MOVED' before it can register 'ad'.

Why it works: It sells a laugh you can aim at a specific person you know — the value isn't 'AI video,' it's a 15-second social payoff (send it to the group chat) with a one-photo cost of entry, so perceived effort is near zero while perceived reward is a named reaction from a named person.

Adaptation: Don't clone their dance clips or their song — clone the STRUCTURE: a held still frame that erupts into synced motion on a downbeat, then a one-thumb mechanic beat, then a variety montage. Build our version around subjects we own outright (consented team photos, licensed stock portraits, our own pets, obviously-synthetic AI-generated 'people') and around a motion style we pick fresh each cycle — dance is one lane, but the same skeleton works for 'make any photo walk toward camera,' 'make any photo talk,' or the current month's motion meme, which lets us rotate before fatigue instead of after. Treat the first 0.8s of stillness as the non-negotiable asset: that's the whole hook, and it costs nothing to produce.

Risks: Likeness is the sharp edge: animating a recognizable real person's photo without consent — especially into a body they didn't perform — is a right-of-publicity exposure in several US states and a straight policy violation on Meta for public figures; use only consented, licensed, or fully synthetic subjects, and never depict a named celebrity or a competitor's founder as the 'anyone.' Music is the second edge: the trend's pull comes from a specific track, and the obvious move (lift the trending song) is the one that gets the ad muted, pulled, or the account struck — license it or use generated/production music that hits the same BPM and downbeat. Third, the demo itself makes an implicit claim: if the ad shows output quality the product can't hit from a single average user photo, that's a misleading-advertising and refund/chargeback problem, not just a creative one. Also avoid the deceased-relative angle even though it converts — Meta enforces on it and the brand damage is asymmetric. Do not reuse their audio, their subjects, their overlay wording, or their exact end-card layout; the structure is unprotectable, the assets are not.

Lineage tag: vgai-c1286868600312178 ← put this in the shipped ad's name/UTM

packet.json ↗

APPROVED The Photo That Never Existed AI Video Generator · 111d · ↻10 · ✂2

Hook: A stack of ordinary phone snapshots — different rooms, different years, different lighting — slides into frame and collapses into ONE studio-grade black & white family portrait that could never have been taken; the gasp is realizing everyone is finally in the same frame.

Why it works: It sells an emotionally impossible object — a portrait of people who were never in the same room — so the viewer isn't buying a filter, they're buying a memory they were denied, and black & white does the heavy lifting by coding the output as heirloom/permanent rather than as a disposable AI toy.

Adaptation: Brief: shoot the ad as a first-person camera-roll POV — a real thumb scrolling real-looking mismatched snapshots, three taps, one generate gesture, then a 2-second color-drain transform into a single composed black & white group portrait held in-hand. Vid Gen AI owns the transformation MOMENT rather than the output style: build our own seeds (cast our own 'family' from licensed/synthetic talent, our own room, our own portrait composition) and differentiate the payoff by showing the source thumbnails ghosting into their positions in the final frame, so the mechanism is legible in under two seconds. Ship the same skeleton across 4+ emotional variants — heirloom B&W, warm-color holiday, wedding, first-baby — since the DCO count on the original says the skeleton, not the palette, is what scales.

Risks: Likeness: any real face in source or output needs explicit release — 'reunite with a relative who's passed' framing invites uploads of deceased persons and third parties who never consented, and depicting real deceased individuals is a defamation/right-of-publicity exposure in several US states. Meta policy: second-person personal-attribute copy ('your family', implying family status/loss) can be rejected under the personal attributes rule — write in third person or product-first. Sensitive-emotion/grief framing risks the exploitative-content policy; keep it celebratory, not bereavement-baiting. Trademark: don't name or visually quote the competitor app, its UI, its icon, or its onelink domain; the B&W portrait aesthetic itself is unprotectable but a shot-for-shot recreation of their edit is copyable-expression risk — rebuild the beats, never the frames. Claims: 'instantly' is a performance claim — match it to real render time or drop it. Also verify generated faces don't resemble identifiable public figures before shipping.

Lineage tag: vgai-c1003409312522810 ← put this in the shipped ad's name/UTM

Verdict note: CB in-session: do this one

packet.json ↗