> For the complete documentation index, see [llms.txt](https://docs.elevateab.com/elevate-helpcenter/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.elevateab.com/elevate-helpcenter/product-image-testing/create-a-product-image-experiment.md).

# Create a Product Image Experiment

Step by step walkthrough for creating a product image experiment in Elevate A/B Testing, including product selection, image replacement, reordering, and AI generation.

***

Six steps, from selecting products to launching.

**Before you start:** the theme extension enabled, and your alternative images ready to upload or already in your Shopify media library.

Open **Experiments**, select **New Experiment**, then **Product Image Experiment**.

### Step 1: General Info

Enter an **Experiment name**. **Description**, **Hypothesis**, and **Assets** are optional.

Under **What do you want to measure?**, choose your goal. **Revenue per visitor** is recommended.

Add-to-cart rate is worth considering for image experiments specifically, because images act early in the decision and add to cart gives you a faster signal. Watch revenue alongside it either way. A more flattering image that lifts add to cart and raises returns is not a win.

Select **Browse all metrics** for the full list, or **Add Custom Metric** to track something else. See Custom Events.

### Step 2: Select Products

Choose which products the experiment applies to, up to **500 products**.

**Search** by product title or ID.

**Filter** by product type, status, collection, or tag. Status defaults to **Active** and also offers All, Unlisted, Draft, and Archived.

**Select All** selects everything matching your current filters. Combine it with a collection or tag filter to test a whole category at once.

**Upload CSV** imports a product list.

Testing a group of products gives you a faster result than testing one, provided the image change is the same idea across all of them. Testing 500 unrelated products with unrelated image changes gives you an average that means nothing.

### Step 3: Configure Product Images

Choose the **Select the number of Variations** dropdown, then set the images for each variation.

Each selected product has its own section. Within it:

**Control** shows the product's current live images. Nothing to configure.

**Variation** starts as a copy of the control. For each image you can:

| Action           | How                                                                    |
| ---------------- | ---------------------------------------------------------------------- |
| Replace an image | Select **Replace Image** on that image                                 |
| Reorder images   | Drag an image by its handle. The first image is the **Default Image**. |

Each product is configured independently. Changing one product's images does not affect another.

#### Replacing an image

**Replace Image** opens **Change Product Image** with three tabs:

| Tab                       | Use it to                                                                      |
| ------------------------- | ------------------------------------------------------------------------------ |
| **Upload from my device** | Upload a new image file                                                        |
| **Select from shopify**   | Pick an image already in your Shopify media library. Search by filename or ID. |
| **Generate with AI**      | Describe the image you want and generate it from the existing product image    |

Select the image, then **Submit**.

**Generate with AI** takes the selected product image and a description of what you want instead. Review what comes back before submitting. An AI generated image that misrepresents the product is a returns problem, not just a testing one.

### Step 4: Traffic Allocation

Set the split with the slider. Even is the default.

**Isolate traffic for this experiment** is optional and off by default. See Traffic Isolation.

### Step 5: Audiences

Optionally narrow who enters, by device type, visitor type, UTM parameters, traffic source, or country. Select **Create Custom Audience** for anything more specific.

Mobile only targeting is worth considering for image experiments, since the gallery takes up most of the screen on a phone and the effect is usually larger there. Only do it if that is your hypothesis, not as a default. See Audience Targeting.

### Step 6: Review Experiment

Check the setup, paying particular attention to which images are assigned to which variation on which product. Then select **Create Preview Experiment**.

The experiment moves into **Preview**. QA it, then launch. See Preview and QA a Product Image Experiment.

### Next steps

* Preview and QA a Product Image Experiment
* Troubleshooting Product Image Experiments