AI video ad guidesJewelry ads

How to make hook variations for video ads with AI

Use AI to generate different reasons to keep watching, not the same opening line in new words. Give a text model your product, audience and strongest proof, ask for a first shot, spoken line and on-screen text per angle, then build each opening as a short clip and swap only those first seconds in an otherwise identical ad.

Updated November 20, 20267 steps6 prompts3,871 real ads

Make my video ads →

$1 today. Cancel anytime before day 7.

The steps
  1. 1Gather the proof you can showWrite down one demo, one product feature and one real customer quote. AI can only build honest hooks from facts you give it.
  2. 2Choose six distinct anglesUse problem, desired outcome, product demo, objection, curiosity and customer proof. Two hooks per angle gives you 12 candidates.
  3. 3Generate hooks with a structured promptAsk for a first shot, spoken line and on-screen text for every hook. Tell the model to flag missing proof instead of inventing it.
  4. 4Cut to three different hooksRemove hooks that share the same idea. Keep three that would make a stranger stop for three different reasons.
  5. 5Build each opening shotGenerate 4 to 5 second vertical 9:16 clips with an AI video tool, ideally from your product photo. Add text in your editor, not in the model.
  6. 6Swap only the first secondsAttach each hook to the same body, offer and call to action. Label the files so you know which hook won.
  7. 7Read retention, then conversionsCompare early retention to find what stops the scroll. Then check clicks or conversions to see if that attention pays.
On this page
  1. The steps
  2. Real jewelry video ads
  3. What a real hook variation is
  4. What long-running jewelry openers teach
  5. Write the hooks with one structured prompt
  6. Turn each hook into a first shot with AI video
  7. Keep everything after the hook identical
  8. Test the angle, then polish the words
  9. Prompts to copy
  10. Six hook angles and what to check first
  11. Mistakes to avoid
  12. Questions

Real jewelry video ads to model yours on

3,871 video ads from 479 advertisers in the Meta Ad Library, read October 5, 2026. The median one ran 39 days, and an ad still paid for after months is the closest public sign that it works. Live ads first.

  • Ran 6 days
  • Ran 1 day
  • Ran 8 days
  • Ran 1 day
  • Ran 10 days
  • Ran 5 days
  • Ran 2 days
  • +3,859 more jewelry video ads

    Hooks, landing pages and run length for each

  • Ran 5 days
  • Ran 9 days
  • Ran under a day
  • Ran 11 days

Every creative, with its copy, format and run length

The openers that ran longest

  • Personalized Wooden Horse Door SignCallie · running 1,123 days
  • Custom 10 on 10 VVS Moissanite Grillz getting made in-house 💎Moissanite Grillz Shop · running 873 days
  • (Set of 10pcs)Personalized Leather Tags for Knitting & Crochet, Handmade Labels for Knitting Items, Gifts for…Insgifts · running 760 days
  • Le cadeau de Noël le plus personnelLoanya France · ran 701 days

What a real hook variation is

A hook variation is a different reason to keep watching, placed in the first three seconds. It is not the same sentence with new adjectives. If two openers both say 'Meet our new necklace,' you have one idea and testing them teaches you nothing. A real variation changes what the viewer sees, feels or expects next.

Every hook has three parts: the first shot, the spoken line and the on-screen text. Write them together. Many people watch with the sound off, so the picture and the text must carry the idea alone. The spoken line should add to them, not repeat them word for word.

AI helps in two places. A text model produces many angles in minutes. A video model produces the opening shots without a shoot. You still choose the angles, supply the proof and decide what ships. Treat AI as a fast drafting partner, not as the person who knows your customer.

What long-running jewelry openers teach

In the Meta Ad Library, the jewelry category shows 3871 video ads from 457 advertisers, with a median run of 46 days. The ads that ran far longer than that are worth studying. FabuLove opened with "Imagine her reaction… Surprise her with a rose preserved to last forever & personalized necklace! 🌹" and ran 611 days. Indigo Lune opened with "3 For 2 Ends Soon." and ran 474 days.

Two more: Lainly's "Buy 1 Get 2 FREE Today Only" ran 383 days, and Brilliant Earth's "Design Your Own Ring" ran 382 days. All four used the Shop now button. Notice that none opens with a brand story. One sells a feeling (her reaction), two lead with the offer, and one names an action you can take. Those are three different hook angles.

Long run time shows an advertiser kept paying, not that the hook alone did the work. Still, it is a useful map of angles that survive. Use the Sprites ads library to see real ads from a category, pull 20 openers from yours, and sort them into emotion, offer and action. Then write one hook in each bucket. For jewelry-specific formats, see jewelry video ads.

  • Emotion: picture the recipient's reaction before the product appears
  • Offer: lead with the deal in the first words, such as 3 for 2 or buy 1 get 2
  • Action: name what the viewer can do, such as design your own ring
  • Each opener is short enough to read in one glance

Want jewelry video ads like these for your product?

The jewelry video ads that kept running, read from the ad libraries. Every change waits for your approval.

Write the hooks with one structured prompt

Start with the facts. Fill in your product, audience, benefit, proof you can show, and placement with tone. The proof line matters most. A demo, a feature or a real customer quote gives the model something true to build on. Without it, you get confident-sounding claims you cannot back up.

Ask for two hooks per angle across problem, desired outcome, product demo, objection, curiosity and customer proof. Require three fields per hook: first shot, spoken line and on-screen text. Add two rules: the opening must make sense in three seconds with the sound off, and the model must use only your facts. If an angle needs proof you did not supply, it should say what is missing.

Then edit hard. Cut any pair of hooks that share one idea. Read each spoken line aloud and keep it under about 10 words. Make sure the on-screen text is 2 to 5 words, since long text gets skipped. You want three survivors that differ in the reason to watch, not in tone. The first prompt below is ready to paste.

  • Give real proof, not just benefits
  • Force the three-field format for every hook
  • Tell the model to flag gaps instead of inventing
  • Pick three hooks from three different angles

Turn each hook into a first shot with AI video

Generate only the opening, not the whole ad. Make 4 to 5 second clips in vertical 9:16 (1080 by 1920). Describe one subject, one camera move and one ending frame. If your tool accepts a reference image, upload a clean product photo on a plain background so the piece looks like your real item. Make 3 or 4 takes per hook and keep the best.

Keep the visual world consistent across hooks: same lighting style, same color palette, same product. Otherwise a test measures production differences, not hook differences. Avoid faces and brand marks in generated clips. Hands, close-ups and macro shots tend to look cleaner and sidestep many uncanny results.

Do not ask the model to render your on-screen text. Generated lettering is often garbled. Add text in your editor, kept inside the safe area so buttons and captions do not cover it. For a real product demo, such as a ring turning in the light, use your own footage if you have it. See the video ads guide for the wider workflow.

Keep everything after the hook identical

The test only works if the hook is the only change. Build one body and reuse it. For a 20-second ad, a simple structure is: 0 to 3 seconds hook, 3 to 10 seconds proof or product in use, 10 to 16 seconds offer, 16 to 20 seconds call to action. Keep the same voiceover, music bed, offer and button across every version.

Make each hook end on a frame that flows into the body. A product-centered final frame works well because the proof segment can start from it. If one hook ends on a hand and another on a wide shot, the cut will feel different and you will be testing transitions too.

Label files by angle, for example Hook_Problem_A, Hook_Offer_B, Hook_Gift_C. Keep a sheet that lists the spoken line and on-screen text for each. Check placement rules in the Meta ad specs so your text sits clear of the interface. Consistent naming is what lets you read results a week later.

  • One body, three hooks
  • Same voice, music, offer and call to action
  • Hooks end on a frame that matches the start of the body
  • File names that include the angle

Test the angle, then polish the words

Run your three hooks as three versions of the same ad with the same audience and similar budget. Compare early retention first, such as how many people stay past the opening seconds. That shows which hook stops the scroll. Then check clicks or conversions. A hook that earns attention but no clicks may promise something the body never delivers.

Decide on the angle before the wording. If the gift-reveal angle beats the offer angle, do not spend your next round rewriting the offer line. Make three new hooks inside the winning angle: a different first shot, a different recipient, a different moment of reaction. Only after an angle wins should you test line-level tweaks.

Run on a rhythm. A weekly round of three new hooks beats one big batch you never review. Keep a log of the angle, the result and what you changed. If you would rather not run the loop yourself, Director by Sprites is a full creative team for your startup. Every Monday you get finished, on-brand video ads, reviewed by a human. Learn more at Director.

Prompts to copy

Paste each one as written, then change the product, the setting and the light to match your own footage.

  1. Text model: 12 hooks, 6 angles

    Create 12 opening hooks for a 20-second vertical 9:16 video ad. Product: [what you sell]. Audience: [who it is for and what they struggle with]. Benefit: [what it helps them do]. Proof I can show: [demo, feature, or real customer quote]. Placement and tone: [e.g., Instagram Reels, casual and direct]. Write two hooks each for problem, desired outcome, product demo, objection, curiosity, and customer proof. For each, give first shot, spoken line, and on-screen text. It must make sense in three seconds with sound off. Use only the facts I provide. If proof is missing, say what is missing.
  2. Hook: problem (tangled chain)

    Vertical 9:16, 4 seconds, close handheld-style shot. A pair of hands lifts a tangled gold chain necklace out of a crowded jewelry dish, fingers working to loosen the knot. Soft window light, warm neutral background, shallow depth of field. Slow push-in. No faces, no text, no logos. Ends on the tangled chain held up toward the camera.
  3. Hook: desired outcome (necklace on)

    Vertical 9:16, 4 seconds. Close shot of a hand fastening a delicate gold pendant necklace at the collarbone, framed from the chin down so no face is visible. Soft daylight, cream linen top, gentle shimmer on the pendant. Slow handheld drift. Ends on the pendant resting still. No text, no logos, no brand marks.
  4. Hook: product demo (ring in light)

    Vertical 9:16, 5 seconds, top-down shot on a matte beige surface. Hands open a plain unmarked ring box and rotate a solitaire-style ring toward the camera so light catches the stone. Clean studio lighting, sharp focus, smooth motion. Ends with the ring centered in frame. No text, no logos, no people beyond hands.
  5. Hook: objection (is it too delicate?)

    Vertical 9:16, 4 seconds. Macro shot of a hand gently tugging a thin sterling silver chain bracelet to show it holds, then letting it swing. Neutral gray backdrop, crisp light, shallow focus. Ends on the bracelet resting on a wrist. No faces, no text, no logos, no brand marks.
  6. Hook: curiosity (gift reveal)

    Vertical 9:16, 4 seconds. A gift box on a kitchen table; a hand slides the lid off to reveal a personalized name necklace on a velvet card, with the lettering blurred and unreadable. Warm evening light, slow push-in, soft background. Ends on the necklace in sharp focus. No faces, no text, no logos.

Six hook angles and what to check first

AngleOpening shotOn-screen text (2 to 5 words)Check first
ProblemHands fighting a tangled chainTangled again?Early retention
Desired outcomePendant being fastened, chin downReady in secondsEarly retention, then clicks
Product demoRing rotating in the lightSee it sparkleClicks to product page
ObjectionBracelet gently tugged to show strengthNot too delicateClick-through and comments
CuriosityGift box lid sliding openWait for itRetention past the hook
Customer proofReal review text over product close-upReal customer wordsConversions

Mistakes to avoid

  • Rewording one idea twelve ways and calling it twelve hooks.
  • Letting the model invent proof, reviews or numbers you cannot back up.
  • Changing the offer, voiceover or call to action along with the hook, which muddies the test.
  • Asking the video model to render text, then shipping garbled lettering.
  • Making hooks that only work with sound on.
  • Judging by clicks alone and ignoring whether the first seconds held attention.

Rather have the videos made for you?

Director by Sprites is a full creative team for your startup. Every Monday you get finished, on-brand video ads, reviewed by a human. You never write a prompt.

Apply to Director by Sprites →

Questions

How many hook variations should I make for one video ad?

Generate around 12, then test three. Twelve gives you spread across six angles. Three is enough to compare without splitting budget too thin.

Can AI write the whole video ad, not just the hook?

It can draft the script and generate clips, but the hook deserves the most attention. Keep the body fixed and swap only the first few seconds so you can see what stopped the scroll.

What is the best prompt for generating video ad hooks?

One that names your product, audience, benefit and provable proof, asks for two hooks per angle, and requires a first shot, spoken line and on-screen text. Add a rule to use only your facts and to flag missing proof.

What aspect ratio should hook clips be?

Use vertical 9:16, 1080 by 1920, for Reels, TikTok and Shorts. Keep text away from the edges so buttons and captions do not cover it. Check the placement guide for each platform.

How do I know which hook variation won?

Compare early retention across the versions to see which opening held viewers. Then look at clicks or conversions to confirm that attention led to action. A hook that wins only on retention is not yet a winner.

Should I use real people in AI-generated hooks?

For clean results, favor hands, close-ups and product shots over faces. Avoid real people's likenesses and brand logos in generated clips. If you want a human face, use footage you shot or have rights to use.

Keep going

Get jewelry video ads made for your brand

  • The jewelry video ads that kept running, read from the ad libraries
  • Finished, on-brand video ads, reviewed by a human
  • Every change waits for your approval