Generate a hook pack before the video exists

The standard production order is: make the video, then write a hook for it. That order guarantees a weak opening, because by the time you're writing the hook, every constraint that matters has already been decided. The pacing is set. The first shot is whatever the first shot happened to be. The promise the hook makes has to be one the existing body can already pay off, which means you're not choosing the strongest promise, you're choosing the strongest promise that happens to be compatible with a video you already made.

Reverse it. Generate the hooks first, as a batch, pick the winner cold, and let that winner dictate what the body has to do. The opening stops being a lid you screw on at the end and starts being the brief.

What a hook pack actually is

create_hook_pack takes a description of your brand and produces a pack of distinct UGC hooks in one operation. Three properties of it matter for the production order argument:

They're planned to be distinct. The pack isn't N independent generations that happen to run together — the agent plans N different hooks so they don't repeat each other. That's the difference between a real test set and four rewordings of the same idea, and it's the property that makes cold selection meaningful.

It charges once for the whole pack, not per hook. The price is shown before you confirm, and it doesn't scale with how many hooks you asked for in the way per-generation billing would. Which means the marginal cost of asking for five instead of three is not the thing that should decide the count.

The output is finished, not raw. Every clip is submitted as a 1080p portrait job rendered with Vidu Q3, and on completion each hook line is burned onto the clip as big TikTok-style text. You are looking at candidates in roughly the form they'd be seen in, not at scripts you have to imagine.

Creating a branded hook pack is the capability page, and the input surface is smaller than you'd expect: a brand description, a count, and optional direction on vibes, emotions, and how the caption is handled.

Brief the pack, not the hooks

The instinct is to write the hooks yourself and ask for them to be rendered. Don't. Writing them yourself removes the one thing you're actually buying, which is a set of options that aren't all downstream of the same idea you already had.

Brief the space instead:

"Give me 5 hooks for my skincare brand — Gen Z audience, playful tone."

That's the whole input. What you add beyond it should be constraints on the space, not contents of it:

What to specify	Why
Brand context	The only required input, and the one that determines everything downstream
Count	How wide a test you want. Three for a quick read, five or more for a real one
Vibes	Constrains tone without prescribing the line
Emotions	Constrains the performance, which is often what actually differentiates two similar lines
Caption handling	Whether the burned-in text is the planned line or your own copy

Vibes and emotions are the underrated pair. Two hooks with near-identical wording delivered at different emotional registers are genuinely different creative, and testing across emotion is often more informative than testing across wording. Versely's own hook library is organised on the same axes — browse the format library and the emotion library and note which combinations you haven't tried.

The selection criterion

Here's where most hook-first workflows fall over. You get five hooks, they're all decent, and you pick the one you like. Personal preference is a terrible selector, because your taste is calibrated on your own brand's existing content and the hook's job is to stop a stranger.

Use a criterion instead, applied in this order:

1. Does it make a specific promise? Not "is it exciting" — is there a concrete thing the viewer now expects to see? "I finally figured it out" promises a reveal. "This is nice" promises nothing. A hook with no promise gives the body nothing to pay off, so it can't drive the edit.
2. Can you pay that promise off in the first five seconds of the body? This eliminates the flashiest option surprisingly often. A hook promising something you can only deliver at second twenty loses the viewer at second six. Reject any promise you'd have to defer.
3. Does it read at thumbnail size with the sound off? The line is burned on as large text, which helps, but a long line rendered large still becomes a wall. Glance at the frame and see whether the promise survives.
4. Is it different from your last winner? Creative fatigue operates on the opening more than the body. If your last three winners shared a structure, break the tie toward the outlier.

Score one through three as pass/fail rather than out of ten, then break ties with four. Typically that leaves one or two out of five, which is the correct yield from a test set, not a sign the pack underperformed.

Now the winner drives the edit

This is the payoff for reversing the order, and it's concrete rather than philosophical. Once a hook is picked, four things about the body are decided that you'd otherwise be guessing at:

• The first line of the body has to connect to the promise, so the script's opening isn't a blank page any more.
• The time budget is set by criterion two. If the promise pays off at second four, the structure is built backwards from that.
• The visual continuity is set by what the winning hook looks like. If it's a home kitchen, a body that opens in a studio is a cut the viewer feels.
• The format is largely decided, since the hook is a 1080p portrait clip and the body needs to match rather than be reframed to fit.

None of those are decisions you can make well before a hook exists. Hook rate is the metric this order optimises for, and it's measured on the opening, not the average.

Assembling it

The hook clip and the body are separate assets that need joining, and joining happens in the editor rather than inside the hook tool.

The pack gives you a finished vertical clip with text burned in; the body is whatever you generate or shoot. Putting them together is a normal assembly: both clips into the editor, in order, with whatever transition and audio the piece needs. You can check the join with a 480p preview at no credit cost — that preview pass carries a short per-user cooldown, and the final export is charged once regardless of clip count. Adding a hook to your video covers the join, and previews and exports covers the billing shape.

There's a second source of openings worth knowing about. The AI Hooks catalog carries premade examples, reaction B-roll, and premium marketplace clips. Clips already in your library — whether they came with the catalog or you purchased them — can be reused without buying them again, while new marketplace clips are purchased separately, and premium previews may carry watermarks until unlocked. Applying any of them is the same editor operation as applying a generated hook. Searching stock UGC B-roll clips covers that path, and it's the right one when you want a reaction opener rather than a spoken promise.

The production order, end to end

1. Brief the pack. Brand context, a count, optional vibes and emotions. One charge, shown before you confirm.
2. Select cold. Apply the four criteria before you've written a word of the body. Keep one, note the alternate.
3. Write the body from the promise. The winning hook's promise is the brief. Time-budget the payoff.
4. Generate the body. Match the hook's setting and format so the join doesn't read as a cut between two different videos.
5. Assemble and preview. Hook first, body second, check the join at 480p before exporting.
6. Bank the losers. A hook that failed criterion two against this body may pass against a different one. They don't expire.

That last step is why the pack is worth generating wider than you need. Four unused hooks are not waste, they're a queue, and the next piece starts at step two instead of step one.

FAQ

How many hooks should I ask for in one pack?

Enough that the winner is a genuine selection rather than a default. Three is a quick read; five gives the criteria something to reject. Since the pack is charged once rather than per hook, let the count follow how wide a test you want.

What if none of the hooks are good?

Look at the brand context you gave it before regenerating. A pack that comes back generic almost always traces to a brand description that was generic, since that's the required input everything else is planned from. Adding a specific audience, tone and product benefit changes the output far more reliably than asking again with the same brief.

Can I use my own hook line and just have it rendered?

Yes — the caption handling lets you supply your own text rather than using the planned line. That's the right choice when the line is already validated and you want it produced consistently. It's the wrong choice when you're still looking for a winner, because supplying the copy collapses the pack back into one idea rendered five ways.

Does going hook-first work for anything other than UGC ads?

The order generalises anywhere the first three seconds decide whether the rest gets watched, which covers most short-form. It generalises less well to formats where the viewer has already committed before playback — a video on a landing page below a headline that's doing the hook's job, for instance. There, the opening's job is continuity rather than capture, and building the body first is defensible.