How do you generate multiple ad hook variations from a single competitor ad?

Upload the competitor ad to the invideo agent: it breaks the ad down shot by shot, explains why it converted, isolates the hook, and generates 7–8 hook directions per iteration — each with a rationale. You then expand the strongest directions into script and voiceover variants, produce only the hook segment while keeping the rest of the ad constant, and ship every variant to let performance data pick the winner.

Start by pulling the winning ad from Meta Ad Library and uploading it to the invideo agent — invideo is an agentic video creation platform that reads uploaded reference videos and holds your brand context across every generation. The invideo agent deconstructs the ad at the structural level: shots, framing, voiceover, setting, cast, and pacing, with timestamps per cut. In one documented run it broke an 11-second reference ad into 7 distinct shots; in another it detected 9 cuts automatically and extracted one frame per scene. That analysis replaces roughly an hour of manual breakdown work before you brief anything.

Before requesting variations, have the invideo agent explain why the ad worked and separate what must stay from what should be swapped. Generic AI tools can't read why an ad converted, so they can't tell which elements are safe to change — load your brand guidelines, product images, and tone of voice into the invideo agent's context first so every variation comes out brand-native rather than a surface copy.

Now ask for hook variations. The invideo agent generates 7–8 hook directions per iteration, each with a rationale for why it should stop the scroll — run additional iterations if you want a deeper pool. Winning UGC-style ads typically follow a four-beat structure — hook, product reveal, value sell, CTA — so treat the hook beat as the swap variable and hold the other three beats constant. That isolation is what makes the subsequent A/B test valid: one changed variable per variant.

Once a hook direction is locked, expand it into full script options. Brief the invideo agent to write three distinct script variations in different tonal registers — an enthusiastic reaction, a quiet understated take, a skeptic angle — then generate three voiceover takes for the chosen script. In one documented UGC production, the invideo agent produced all three script directions in a single pass after the hook was locked.

Produce the hook clips themselves with two tactics depending on the hook type. For hooks built around a physical action (a snap that cuts the location, a reveal gesture), generate the exact first frame as a still image, compose it precisely, then animate from that locked keyframe — text descriptions of complex actions are unreliable. For hooks that depend on precise timing or pacing, upload a reference video of the feel you want instead of describing it in text; the invideo agent reads timing and rhythm from video far better than from prose, and routes each clip to the right video model — Kling, Veo, or Seedance 2.0 — without you switching tools.

Concentrate your iteration budget on the opening seconds. The first 3 seconds decide a UGC ad's performance, and the scroll-stopping window is roughly 1.5 seconds — so spend the majority of production time regenerating and refining hook clips, not the body of the ad.

Finally, ship every variant instead of pre-selecting one. You can't predict which hook wins without running it, and the economics support volume: documented runs land at roughly $75 and about an hour per finished ad, with 8–10 variations of a single winning ad achievable in one day. Scale whichever hook the ROAS data validates, then re-upload that winner to the invideo agent to generate the next round of structural variants for new audiences, seasons, or platforms.

Watch some of these to see what works for you:

It watches the ad, breaks it down shot by shot, figures out what made it work, and rebuilds the whole thing. Storyboard, shot list, generations, final edit.

— invideo's creative team, describing the competitor ad replication workflow