VIBECODEIUM

Product ad variations that you can actually test

Turn one offer into three angles and a small set of image and copy variations, sized for each placement and named so results are readable.

AdsIntermediateAbout 45 minutes

Connected glass modules linked by copper pipes in a dark studio

What you will make

Nine ad variations (three angles × three formats) with matching copy and a naming scheme, ready to run as a small test.

Best for: Sellers and small agencies who run paid social and need more creative variations than they can shoot.

Before you start

  • One product, one offer and one audience. Do not start with three of each.
  • A product photo (or the layered file from the product photography guide).
  • Access to the ad platform’s current creative specs and policies.

Steps

  1. Write the offer in one sentence

    Product, who it is for, what changes for them, and any real offer (price, shipping, guarantee). If you cannot say it in one sentence, the ad will not either.

  2. Pick three angles

    Common angles are problem/solution, social proof you actually have, and a use-case demonstration. Do not invent testimonials or statistics. If you do not have proof yet, use demonstration and problem/solution.

  3. Draft hooks before visuals

    For each angle write five opening lines of under ten words. The first line of copy or the first second of video decides whether anyone keeps looking, so choose the visual after you know the hook.

  4. Generate the visual per angle

    Keep the real product exact (see the product photography guide: generate the scene, composite the real photo). Make one visual per angle, then adapt it to 1:1, 4:5 and 9:16.

  5. Respect safe zones

    Platform interface elements cover the top and bottom of vertical placements. Keep text and logos away from the edges and check each platform’s current safe-zone guidance before finalising.

  6. Name everything

    Use one scheme, such as product_angle_format_hookNumber_v1. When results arrive you want to know which angle and which hook won without opening each file.

  7. Run a small test and read it honestly

    Give each variation a small, equal budget and decide in advance what result counts as a win. Change one thing at a time. Do not call a difference real if it is a handful of clicks.

Prompts

Replace anything in [square brackets]. Prompts are starting points: results vary between tools and versions, so review every output.

Angles and hooks

Use this with a text model after you write the offer sentence.

Offer: [one sentence]. Audience: [who]. Write three angles (problem/solution, use-case demonstration, and one more that does not require customer reviews I do not have). For each angle write five hooks under ten words. Do not invent statistics, testimonials, prices or guarantees. Mark any claim that would need proof.

Visual brief

Use this to describe the scene for one angle.

A [scene that demonstrates the angle], clean modern composition with clear empty space for a headline at the [top / left], natural light, shallow depth of field, no text, no logos, no packaging in the scene (the product will be added from a real photo).

Policy self-check

Use this before submitting ads for review.

Review these ad texts for claims that could breach common advertising policies: personal attributes, before/after health claims, guarantees of results, misleading urgency, and unsupported comparisons. List each risky phrase and a safer rewrite. Texts: [paste]. I will still check the platform’s current policy myself.

Checklist before you publish

  • One offer, one audience, three clearly different angles.
  • No invented reviews, numbers, awards or guarantees.
  • The product in every image is the real product, accurately shown.
  • Text and logos sit inside the safe zone of each format.
  • File names follow one scheme.
  • You checked the platform’s current creative specs and policies.

Common mistakes

  • Testing three angles and three audiences at once.
  • Letting AI imagery show features the product does not have.
  • Declaring a winner from a tiny sample.
  • Copy that promises results you cannot document.

Automate the next step

Related guides