Generate A/B test ad variants at scale with AI

A fashion brand running paid social campaigns generates two distinct ad image variants from a single product photo in one pipeline run. Nano Banana Pro writes the ad concept and copy. GPT Image renders both variants. The full month of A/B test assets takes minutes, not days of manual assembly.

Why is producing A/B test ad variants manually so slow?

A/B testing ad creative requires volume. Two variants per product, across five products, updated monthly, means 120 distinct ad images per year before accounting for platform-specific crops and format variations. Each image typically requires a brief, a designer, a round of feedback, and an export. That process does not scale.

Most teams end up running fewer tests than they should because production is the bottleneck, not ideas. The result is campaigns that run on untested creative for longer than is optimal.

How do brands generate multiple ad variants from one product photo?

The approach this brand uses starts with a single product image and two text inputs: a description of the product and a brief for the ad campaign. Nano Banana Pro generates the creative concept and ad copy for each variant. GPT Image renders both variants as finished ad images, each with a different visual treatment, ready for upload to paid social platforms.

The pipeline runs as a batch. A set of five product photos produces ten ad images in a single job. Each image comes out at the dimensions set before the batch runs.

How the cnaps studio pipeline works

This pipeline runs 7 nodes.

Step 01: Image Loader (Input)
The product photo loads into the pipeline. JPEG or PNG.

Step 02: Text Input x2 (Input)
Two text inputs enter the pipeline: a description of the product and a campaign brief. These drive the creative concept generation in the next step.

Step 03: Nano Banana Pro (AI Model)
Nano Banana Pro receives the product image and the text inputs and generates the creative direction and ad copy for each variant. The two variants are given distinct visual and messaging treatments based on the campaign brief.

Step 04: GPT Image (AI Model)
GPT Image renders both ad variants as finished images using the creative direction from Nano Banana Pro. Each variant is a complete ad creative with the product image incorporated into the final composition.

Step 05: Image Viewer x2 (Output)
Both ad variants are displayed for review and download.

What makes this approach faster than working with a designer?

The pipeline removes the brief-to-asset handoff entirely. The creative direction, the copy, and the rendered image all happen inside the same pipeline run. There is no waiting for a designer to pick up the brief, no feedback loop on the first draft, and no manual export step. The output is a finished ad image, not a design file that requires further processing.

For teams running weekly campaign updates, the time saving per cycle is significant. For teams that have wanted to run proper A/B tests but have not had the production capacity to do it consistently, the pipeline makes it operationally viable.

Which products and campaign types work best with this pipeline?

The pipeline works best for products where the visual is the primary selling point: apparel, accessories, beauty products, and home goods. Campaign briefs that specify a clear audience, a tone, and a desired action produce the most distinct variant pairs. Vague briefs produce variants that are too similar to generate meaningful A/B test data.

Try this pipeline on cnaps.ai

Cnaps.ai is a no-code visual platform for building and running multi-model AI pipelines. Fork the A/B test ad variant pipeline from the community and run it against your own product catalogue. No code required.

View and fork this pipeline on cnaps.ai

Frequently asked questions

How different are the two variants from each other?

The degree of variation depends on the campaign brief. A brief that specifies two distinct audiences or two different value propositions produces variants with clearly different visual treatments and copy. A generic brief produces variants that may be too similar for meaningful A/B testing. Specificity in the text input drives variation in the output.

Can the pipeline run more than two variants per product?

The current configuration produces two variants per run. To generate additional variants, run the pipeline again with a modified campaign brief. Each run takes the same product image as input, so the visual asset is reused while the creative direction changes.

What dimensions do the output ad images come out at?

Output dimensions are set before the batch runs. Set the target dimensions to match your platform specs before running: square for Instagram feed, vertical for Stories, horizontal for display. The pipeline produces both variants at the same dimensions in a single run.

Can this pipeline run on a batch of product photos at once?

Yes. In batch mode, cnaps studio processes multiple product images in sequence, producing two ad variants per product. A set of five product photos produces ten ad images in a single batch job.