Dropdown

TLDR

  • AI image generation tools like ChatGPT 4o can help e-commerce brands quickly turn basic product photos into styled marketing assets.
  • The tech isn’t perfect yet (minor distortions, manual steps), it dramatically speeds up creative workflows.
  • Build predefined style prompts for consistent results across your catalog
  • Automate the workflow with Zapier or Airtable to provide a simple interface for your entire team to use

AI Image Generation Is A Game Changer

I’ve been generating images with AI for about a year now, and the practical use case thus far has been fairly limited (outside of sending around funny images to group chats).

Last month, Open AI announced a new image generation model inside of Chat GPT 4o, which immediately catapulted it to the #1 spot in the image generation game.

Is it perfect? No. But with the speed at which they are updating these models, the potential for this to sit inside our e-com marketing workflows is enormous.

Marketers and designers who know how to use these AI tools can 10x their productivity. But we still need a human steering the ship to guide our AI tools to create the best result.

The rest of the article will focus on the following:

1. How to Use AI Image Generation Today

2. Limitations to Be Aware Of

3. Automating the Workflow

Let’s get started.

How to Use AI Image Generation Today

Turn plain product images into ready to go ads, for social and email, in a fraction of the time. Below are the exact steps that I used to create an email campaign image in about 10 minutes.

  1. Start with a plain product image. A studio shot from your photographer works great. If you don’t have one yet, you’ll be amazed at the results from using a photo shot on your phone. As an experiment, I took a photo on my iPhone of my Hydro Flask water bottle. As you can see this photo is objectively terrible.
iPhone photo of my Hydro Flask

2. Using ChatGPT 4o model, upload the image and ask Chat GPT to create a prompt for you based on your objective. Most people skip this step, but the prompt is the key to creating something usable. If you have an example image of a final style that you are aiming for, you can upload that image and ask ChatGPT to describe the style.

My exact prompt below used to create an image for an email campaign:

Ask Chat GPT to give you the ideal prompt

3. Copy the prompt and make any changes if needed. Paste the optimized prompt back into ChatGPT and ask it to render the image. Here is the prompt that was generated for this Hydro Flask example:

Reword the prompt as needed to fit your style

4. Tweak the generated image if needed. Explain what needs to be changed, and ask Chat GPT to create a new image with the adjustment. My result was pretty solid on the first try. The newest image generation model nails the lighting and shadows that are needed for the image to look real.

Initial Render From AI

5. Since I’m using this for an email or social campaign, I need to layer in a headline and call-to-action.

Below I’m stacking up my ad (with headline and CTA made in Canva) against the auto-generated ad with headline and CTA overlay from ChatGPT.

The Chat GPT version is objectively better, with one exception: Chat GPT distorts the product image further to fit the Headline and CTA.

My Canva Edit

Chat GPT Edit

Limitations to Be Aware Of

Product Representation

You may notice right away that the product rendered by AI is 95% accurate, but there are a few inconsistencies when compared to the original image shot on my phone. The GPT rendered image is smaller, with slightly different dimensions than my original photo.

This may be controversial for some brand owners, and may discourage them from using this tech altogether, for fear of misrepresenting their products. However, smart operators will find a use case for this powerful tool. Photographers touch up images prior to delivery, and this sits in a similar gray area of realistic representation.

When was the last time you had a fast food burger that looked 100% like the image you saw on the commercial?

When was the last time your Whopper looked like this?

Fragmented Workflow

As someone who is always looking for the automation angle, I wish the process were more automatable. I have a few ideas on how to make this more automated, but this workflow will work best with some human intervention. It will get there, but for now this should be used to speed up the current asset creation process with a human in the loop. In the hands of a creative designer, this can yield some amazing, fast results.

GPT Hallucinations

Although we got pretty good results right away on the background for this image, there can be a lot of back and forth with the AI to get things “right”. Part of this can be sped up with strategic prompting, but the more you tweak style instructions in a single chat, the more ChatGPT blends your versions together. This can make it difficult to change styles completely in a single chat. I’d recommend starting a fresh chat after narrowing in on your style preferences.

Automating the Workflow

Next steps

  • Work with your marketing team to test this as a manual workflow. Does it save them time and produce an acceptable result?
  • Dial in 2-3 style prompts that can be reused across products.
  • Measure the time savings or increased output

Want help building? We help brands automate workflows like this one.

Get in touch to customize this workflow for your team