Using ChatGPT to write ad copy became increasingly popular in 2023, with brands such as Sephora and Mint Mobile using the AI tool to create campaigns. And while direct-to-consumer brands have been using ChatGPT to write marketing copy too, they’re also exploring what other marketing challenges AI could help them address. 

Take DTC boxed wine brand Juliet Wine, for instance. Allison Luvera, Juliet’s co-founder and CEO, said in an email that she’s used ChatGPT to generate relevant copy ideas and edit captions in Juliet’s tone of voice on social media, paid advertisements and other marketing materials. 

Also read: ChatGPT had an eventful 2023

When it comes to integrating AI into her brand’s marketing strategy, Luvera didn’t stop at ChatGPT. Lately, she’s been testing out AI image generation tools such as Flair.AI in an attempt to create product photography on a budget. 

“Modern brands need to be content powerhouses,” she said, adding that for “early-stage start-ups with limited resources this presents a unique challenge. … How do you create thousands of pieces of quality content each year with small budgets and a lean team?”

Luvera’s not alone: Marketers from DTC brands such as Ipsy, Wisp and Sloomoo also said that they’ve experimented with using image generation tools in an attempt to feed the ever-hungry content beast on the cheap. But these brands haven’t used any AI-generated images in their marketing efforts thus far, because the resulting images are often far from what the brand asked the AI tool to create.

“We were excited to try AI product photography tools like Flair.AI that offer to create low-cost creative assets for e-commerce and social media usage,” Luvera said.  But that excitement quickly dwindled upon closer inspection of the images Flair.AI created. “Our trials revealed that although the images generated by AI looked acceptable at a glance, upon closer inspection they often lacked realistic details,” she said.

Luvera pointed out that when attempting to create an image for Juliet’s boxed orange wine product, one resulting image “resulted in a flame oddly appearing on the package top.” When she asked the tool to create an image of a person holding a wine glass, the tool “consistently produced distorted hands,” she said.

“We think tools like Flair.AI are promising but need some refinement before we will be able to leverage them for our marketing needs,” Luvera said.

According to Luvera, the right AI tool for images could not only save a brand money but also time. “A single photoshoot requires hours of planning, sourcing photographers, budget for models, studio space, lighting, props, editing, etc. AI tools also cost money—though some are free or charge a low monthly fee after a free trial period—but those costs are for production ease and unlimited images, versus the set number of images produced in a single photoshoot,” she said in an email.

For Juliet, Luvera estimates that “an AI tool that delivered strong product images” would save the brand roughly $25,000 a year. For comparison, Flair.AI offers a free trial of 100 images per month and a “pro” version offers unlimited images for $10 a month.

Also read: How AI would script Super Bowl ads

Karen Robinovitz, co-founder of DTC slime brand Sloomoo (which sells slime that both children and adults can play with, similar to the way one would play with Play-Doh), said that she’s experimented with AI image generators such as DALL-E (which, like ChatGPT, is owned by OpenAI) and Midjourney because “image is what not only sells a product, but what captures the eye in the constant swipe culture that we live in.”

The AI image generators cited in this story, including Flair.AI, DALL-E, Midjourney and Wisp did not respond to quests for comment. 

The AI image tools Sloomoo experimented with presented a potential financial upside: These tools were free, whereas a “half-day shoot for product image stills on a flat background” costs about $3,000. Sloomoo conducts those types of shoots “every few months for new collection launches,” Robinovitz said. 

But although Sloomoo has had success using AI in other parts of its business, including using ChatGPT for a bit of help with customer-facing newsletters and written content, as well as assistance creating training modules for employees, the images that the AI created for Sloomoo weren’t usable, Robinovitz said.

The AI-generated images Robinovitz has seen for Sloomoo are more post-apocalyptic than satisfying, she said, joking that she asked the AI for an image of a “joyful sunny day, not a doom and gloom day.” 

For someone to stop scrolling on their social media feed, Robinovitz said, an image “has to look really good … either really ‘hashtag satisfying’ in that dreamy ASMR way, or it has to be really fun and engaging.” 

AI images might be a “you get what you pay for” situation. In order to generate images that are closer to being usable, Robinovitz said, Sloomoo would likely have to pay to bring in “specific talent to guide us” through the process, lessening the financial benefit of the effort.

DTC sexual health brand Wisp also experimented with AI image-generation tools, although Jenny Dwork, Wisp’s VP of brand marketing, declined to share which tools the team used. She noted that Wisp has also looked into using AI for other aspects of its marketing efforts, including using ChatGPT for marketing brainstorms.

Because Wisp is still a startup, Dwork said, it has limited resources—from budget to actual headcount—so it was hoping that AI could fill in some of these gaps. Wisp estimates that AI-generated images could help it save tens of thousands of dollars a year if they were usable.

Like many other DTC brand marketers, Dwork did not find the resulting images usable—for instance, the AI could not generate a remotely realistic-looking image of a see-through bottle of pills. She said in an email that these images were from “a beta test from someone who previously consulted with us,” so they did not cost the brand any money to create.

Some AI image generation tools only allow brands to describe what they want the image to look like with written words, but the one Wisp used let the team train the AI with a set of product images—and even then, they weren’t usable. “You feed the AI with a set of images so it learns about the product—it takes some time to train it, the AI. And then you can give it a specific art direction,” such as “I want an image of this bottle on a nightstand, or with a couple,” and then the AI generates images from there. 

“Like all other areas of AI, it’s not perfect,” Dwork said.