Generative AI has been a key discussion topic all year. Online retailers are already incorporating it into their design processes to come up with new products and variations of existing products.

With artificial intelligence learning how to do an endless variety of tasks, online jewelry manufacturer J’evar decided to develop its own generative AI application to design new products.

The tool allows J’evar’s jewelry designers to input information about the product’s materials and specifications, and the generative AI will produce an image of that product. The tool saves the brand weeks of manual design time on products, says Amish Shah, founder and CEO of the direct-to-consumer brand of jewelry featuring lab-grown diamonds.

J’evar began using its jewelry product AI generator last year in 2022. The retailer feeds metrics and images into a knowledge bank — or a database of text, images and metrics for materials that include the weight of gold and silver, among other key details — for the generator to refer to before it produces an image. Shah jokingly refers to it as “JevarGPT 1.0” and “AI for Jewelry 1.0,” the former a reference to OpenAI consortium’s ChatGPT.

For example, Shah says if he wanted to make a bangle, he could input a text prompt to the generative AI, specifying how much the weight of gold should be for that piece, how thin or wide it should be and what design style he would like. He can even ask it to produce 50 iterations from that single prompt. In return, the generative AI will output complete designs, some of which might be ready to turn into tangible products. Other product designs the AI outputs require J’evar designers to modify the design until it can be producible.

Shah says one key reason J’evar can’t produce all the designs is because of the inability to cut diamonds into the shape the AI generates. But even this ability is coming soon with new machinery, he says.

Exploring generative AI for design

“You’re looking at optimization, efficiency, speed — which is of course going to lead to cost reduction in the longer term,” Shah says about using generative AI. “But importantly, from an output perspective, we’re looking at precision and a higher level of creativity.”

In the past few years, developers have trained artificial intelligence to do more than analyze data and tell its users what to make of the data. They’ve trained AI to generate writing, images, videos and sounds. This is called generative AI, and online retailers have already begun to design new products and produce new variations of existing products — and do so quickly. With generative AI, retailers can create and test multiple product ideas in just minutes, much faster than the weeks or months it might take to design now. Online retailers, including J’evar and Auricle Technology, are learning how to use generative AI to assist their product designers, making the process more efficient. But because generative AI is still new, it has limitations on what it can do.

J’evar uses its own generative AI technology to help its human designers speed up the creative process.

J’evar uses its own generative AI technology to help its human designers speed up the creative process.

Gen AI’s value outshines its current limitations

While generative AI is excellent for learning and processing massive amounts of data, it is not yet at a point that it can understand movement through space, says Brendan Witcher, vice president and principal analyst at research firm Forrester. It doesn’t think about engineering and structural elements or physical viability yet, he says. Although generative AI is not at that point yet, he says, that doesn’t mean it can’t be eventually.

“You want to design a shoe. Great,” Witcher says. “Well, a shoe’s a shoe until you put it on and have to run in it, then it falls apart on you because you didn’t think about the physics of how movement happens.

“The big question is when do we bridge the gap between the work that needs to be put into generative AI to understand the movement through physical space that objects need to go through often, and the commercial viability of doing that.”

Informed assessments

However, even with its current limitations, Witcher says generative AI’s value comes from the assessments it already has learned to make. He says people do their jobs based on the knowledge they receive in their training, and “AI kind of works like that too.” But generative AI “takes it to the next step” and looks at more data than humans can process, and then make assessments about what the best subsequent steps are. It can also come up with ideas humans couldn’t or wouldn’t think of because the human mind doesn’t process information the way artificial intelligence tools can, he says.

“We can’t absorb that much data and extract from it an idea. It’s just impossible for us,” Witcher says. “It shouldn’t be lost that just developing an image of something that you wouldn’t be able to think of because you weren’t trained to think that way has huge possibilities.”

Witcher says generative AI’s value extends beyond production speed to unique creations.

“A lot of people talk about generating imagery with AI, but what to me is most important is the ability to do it over and over and over again until you get something you like,” Witcher says.

Will generative AI replace human designers?

Generative AI is not here to replace humans in the design phase, Shah says. Especially not in the jewelry industry.

“Human intelligence supersedes artificial intelligence, at least I can say that for jewelry,” Shah says.

Generative AI is more like an assistant to human designer, Shah says. It’s not the technology that’s telling designers when a piece has been finalized. It’s a human making that decision, Shah says. Just like Adobe and Corel are graphics software tools for designers, generative AI is a design tool, not a human replacement, he says.

“Once we get the initial output, it is then modified to be producible,” Shah says.

Forrester’s Witcher agrees that AI should be used as a tool and not a replacement for creative individuals.

“If all the people learn how to do things on generative AI, then no one learns how to do it beyond generative AI,” Witcher says. “Over time, you start weeding out the expertise from the low-level individuals and nobody becomes a high-level individual.”

Witcher adds that the majority of AI use isn’t leaving artificial intelligence “to its own devices.”

“It’s more assisted intelligence — the AI standing for assisted intelligence — where we’re using it to be more productive in our own jobs that we currently do today,” Witcher says.

J’evar uses generative AI to speed up the design process

Traditionally, Shah says, jewelry design is a long process that can take a few weeks or even more than a month. In the case of commissioned designs, J’evar designers would first have to understand what kind of product a customer wants before going into iterations. In the example of designing a bangle, the designers would have to first determine if a customer wanted a wide cuff or something they could stack, something lightweight or heavy, thick or thin, if they wanted diamonds or gemstones, and so on.

Then, the designers would do initial mockups to ensure they understood the customer’s request correctly. This process would typically be one to three weeks of showing designs to the customer and sketching accordingly, Shah says.

In one case, Shah says he and his team had gone through 55 variations before a customer said, “I love it.” After that, his team would then go to computer-aided design (CAD). From there, it would go to rendering.

“By using AI, we are able to take that process down to pretty much hours and in some cases, literally within minutes,” Shah says.

Moreover, when working manually, the designer has to move every single diamond into place, making sure they are in the correct position. AI speeds up that process, Shah says. In milliseconds, the generative AI processor can move diamonds and gemstones, raise or lower gold weight, or change the width or thickness.

“It’s almost like putting a thousand designers and the type of work they would have done into the knowledge bank and then letting the system do a combination from those thousand designs to give you back results,” he says.

J’evar fed years’ worth of jewelry data into its generative AI platform. The platform produces images that human designers then adjust in the design phase.

J’evar fed years’ worth of jewelry data into its generative AI platform. The platform produces images that human designers then adjust in the design phase.

Developing a custom generative AI processor

Shah says his family’s 90-year history in the jewelry business gives him an advantage over others in developing custom generative AI technology for J’evar.

“It sounds complex, but you have to keep in mind: We’re in the business,” Shah says. “We’re in the jewelry business, so the core bank or the core information that’s required is sitting with us. It’s not something I have to go outside and source.”

J’evar feeds text and imagery into its generative AI to teach it what to output. When inputting prompts, J’evar designers primarily use text to generate an image.

“That knowledge bank is sitting there,” Shah says. “Now, it’s all about organizing it and feeding it into the system in a format that can then be analyzed and the GPUs can run and start combining things and getting them back to you.”

Iterations at scale

Sometimes, what generative AI produces needs less human modification than others. For example, Auricle Technology uses generative AI tools to swap out logos and colors on its different products.

Auricle Technology founder William Cooksey says he created his electronics accessory brand out of necessity. He uses Apple AirPods for long hours most days, and the hard plastic begins to hurt his ears after a while. That led him to create AirPod skins made from silicone that are softer and anchor better into his ears.

When his manufacturer sent back prototypes, it printed Auricle’s logo on them. That led Cooksey to realize the importance of branding and how he can “pivot and get into the licensing game.”

The direct-to-consumer brand launched in 2021 now creates customized merchandise including AirPod skins, AirPod charging case skins, phone cases, wireless chargers and mouse pads. And through licensing agreements, it prints these products with logos for more than 90 teams in Major League Baseball, the National Hockey League and Major League Soccer, as well as about 130 college teams.

Rather than have a designer manually change the colors and logos for each team, the brand has integrated generative AI into its design process, Cooksey says.

Unlike J’evar, however, Auricle does not have the budget nor the in-house capacity to develop an all-new generative AI engine. Its business model is also different, focusing on customization rather than new product development.

New technology, but make it affordable

Cooksey instead works with Goals Media Group to use generative AI into its product iterations. Goals uses technology Microsoft for Startups provides, says Goals founder and CEO Aubrey Flynn. This means it receives access to Microsoft’s resources, including technology and tech experts, among other benefits. Microsoft has announced it would invest $10 billion into OpenAI — the company behind text-based generative AI brand ChatGPT and image-based DALL-E.

Cooksey says his lead designer and Flynn determined generative AI was the way to go from designing products with one team logo to hundreds “in a short period of time without breaking the bank.”

Auricle also uses Goals and its generative AI offerings to develop marketing materials like images for social media that highlight products from different teams at different stadiums. The generative AI creates an image complete with Auricle branding, the team’s branding, and any copy it needs.

“Being a small business, not having a lot of capital, it’s really exciting me that we can still come up with quality images without breaking our budget,” Cooksey says.

“When you deal with those leagues, they want you to be able to launch all the teams at the same time,” Cooksey says. “I just wouldn’t have been able to afford to do that.”

Generative AI’s impact on metrics

Goals has about 650 clients and nearly half are online retailers, Flynn says.

Flynn says that social media marketing creatives that generative AI produced can increase consumer interactions with the ad by more than 35% compared with that brand’s normal creative, according to data from its clients.

This includes creatives entirely generated through AI, visuals that already existed that AI has augmented, and copy that generative AI has helped develop for those types of visuals.

“I’ve seen AI-powered creative outperform to the extent where cost per click on a certain product may have been 30%-40% less expensive based on some of the guidance from AI on the copy and the imagery,” he says.

He adds that brands like Auricle — which lack access to capital, resources, infrastructure and more that large brands have — need to adopt technology like generative AI early on because it’s less expensive than some alternatives like hiring designers or manufacturers from the start.

Early results are insightful, but ‘is this just another buzzword?’

Shah, Cooksey, Witcher and Flynn all expressed the same idea: It may be early, but the application of generative AI in product development is promising.

While some may say generative AI is another buzzword, Witcher says what separates this technology from other tech fads is that companies are already allowing individuals in their organizations to play with, understand and experiment with generative AI.

“They’re almost crowd-sourcing proof of concept,” Witcher says. “It’s a unique characteristic to generative AI that it’s so easy to do and work with that almost anybody can do it.”

Although results are limited in some ways and sometimes imperfect, online retailers are using generative AI imaging to design new products essentially from scratch, customize existing products and develop marketing content. They can develop multiple iterations of these images at once or continue iterating on the same image multiple times until they’re satisfied with how the image looks. They can then take the design that generative AI produces and tweak it manually, saving them the time of doing each iteration manually — and saving them the creative energy it takes just to design a new product iteration.

Do you rank in our databases?

Submit your data and we’ll see where you fit in our next ranking update.