Deep-fashion, using AI face-swapping for retail

Many retailers are experimenting with generative AI tools like Midjourney, Stable Diffusion and Dall-E 2, but quickly realise some fundamental limitations. 

For example, if you generate four images of a specific product, each of the four will be different. There are methods for training the AI model on your specific image which help greatly, but there will still be variance in the details of the product. Also, generating realistic text (including logos) is challenging with generative AI tools currently.

For retailers, the devil is in the details.

This week I started experimenting with a different type of generative AI, a newly released app called InsightFaceSwap which runs as a bot within Discord (instructions here).

As you would guess from its name, InsightFaceSwap uses AI to swap one face onto another. In my experimentations and feedback from friends on LinkedIn, a number of potential use-cases for retailers arose. This post is a deeper exploration into the retail potential for AI face-swapping.

Face-swap as a virtual mirror

When buying any product online, but in particular clothing and accessories, customers want to visualise themselves in the item. A whopping 24% of online apparel orders are returned, creating an immense burden to both the customer and the retailer.

ImageFaceSwap provides a quick and easy ‘virtual mirror’, for example;

Try on a ski jacket

Check out my new shades

Try a new hair-style

Thinking about joining the gym? Try on some new muscles

Face-swap for product images

Aside from the potential to provide a more personal customer experience, and reduce returns, there are some practical applications for retailers using face-swap technology;

• Imagine you have a brand ambassador, and they are not available for a photo shoot, with their permission you can easily face-swap them into a scene.

• If you have an existing product on your site and are adding a new variant, but the original model is not available. Using face-swap you could ensure consistency of your product photos. But be careful, Levi’s recently sparked controversy with their decision to trial AI Avatar models.

• The same could be done with your current library of photos, where models, looks and hairstyles may have changed over time. A retailer could use face-swap across the back-catalog to provide consistency across your online store.

• It would even be possible to personalise the entire customer experience based on the shopper’s profile. Imagine showing all of your products (perhaps with an opt-in) with models of a similar gender, age, ethnicity and style to the shopper, like this Shopify app does. What impact could this have on conversion rate?

There are many ways face-swapping tools like InsightFaceSwap could be used by retailers, but we believe that generative AI is going to change online retail more broadly. Conversational (chat) commerce, voice interfaces and new personalisation tools are just come of the exciting changes that are on the way.

If you’re interested to discuss the future of online retail when combined with the power of generative AI, please get in touch.

Bot view 🤖

In a new experiment for our Imagine posts, we are going to ask ChatGPT to summarise, challenge and add it’s own thoughts to each article. Here we go!

🤖 tldr: “The article explores the potential of AI face-swapping tool, InsightFaceSwap, in the retail sector, addressing limitations of other generative AI tools. The tool could enhance customer experiences, maintain consistency in product imagery, and offer personalization by letting customers visualize themselves with products or presenting products modeled by similar-looking individuals.”

🤖 counter view: “While the concept of using AI face-swapping technology in retail, such as InsightFaceSwap, certainly seems novel and intriguing, there are several potential issues that warrant consideration. Firstly, the precision and accuracy of AI-generated images may still leave much to be desired. This could lead to a discord between the product as visualized by the customer and the actual product, which could in turn lead to dissatisfaction and increased returns. Additionally, implementing such technology would likely require a considerable investment in terms of both time and resources, potentially outweighing any benefits of decreased return rates or increased customer engagement.

Secondly, there are potential ethical and privacy implications of AI face-swapping technology that are yet to be fully addressed. For instance, creating images where customers can see themselves in different apparel or accessories might require the collection of detailed personal data, which can raise privacy concerns. Furthermore, the idea of personalizing the shopping experience by using models who match the customer's profile may also inadvertently promote harmful beauty standards or stereotypes. Such issues may detract from the overall customer experience, rather than enhancing it. As such, retailers must balance the innovative potential of such tools with their potential risks.”

🤖 Additional thoughts: “The article discusses how AI face-swapping technology, like InsightFaceSwap, could benefit retail by overcoming limitations of other generative AI tools. The technology offers possibilities such as enabling customers to visualize themselves with products, thus enhancing the shopping experience and possibly reducing return rates. It also provides options for retailers to maintain product image consistency and offer personalization by showing products modeled by similar-looking individuals. Despite the potential, the implementation of such technology raises questions about precision, resource investment, and potential ethical and privacy concerns. Ultimately, it emphasizes the need for balance between the innovative potential of such tools and their potential risks.”