It’s been just a few weeks since OpenAI allowed customers to commercially use images created with DALL-E 2, the remarkably powerful AI text-to-image system. But despite the current technical limitations and lack of volume licensing, not to mention API, some pioneers say they are already testing the system for various business use cases – waiting for the day when DALL-E 2 becomes stable enough to be put into production.
Stitch Fix, the online service that uses recommendation algorithms to personalize clothing, says it: experimented with DALL-2 to visualize its products based on specific characteristics such as color, fabric and style. For example, if a Stitch Fix customer requested a “tall, red, stretchy, skinny jeans” during the pilot, DALL-E 2 was tapped to generate images of that item, which a stylist could use to match with a similar product in Stitch Fix’s inventory.
“DALL-E 2 helps us bring out the most informative features of a product in a visual way, ultimately helping stylists find the perfect item that matches what a customer requested in their written feedback,” a spokesperson told e email to londonbusinessblog.com.
A DALL-E 2 generation of Stitch Fix’s pilot. The prompt was: “soft, olive green, great color, pockets, pattern, cute texture, long, cardigan.”
Of course, DALL-E has 2 quirks, some of which give early business users a break. Eric Silberstein, the VP of data science at e-commerce startup Klaviyo, outlines in a blog post his mixed impression of the system as a potential marketing tool.
He notes that facial expressions on human models generated by DALL-E 2 are often inappropriate and that muscles and joints are disproportionate, and that the system does not always understand instructions perfectly. When Silberstein asked DALL-E 2 to create an image of a candle on a wooden table against a gray background, DALL-E 2 sometimes erased the lid of the candle and blended it into the desk, or added an inappropriate border around the candle.
Silberstein’s experiments with DALL-E 2 for product visualization.
“For photos with people and photos of people modeling products, it cannot be used as it is,” Silberstein wrote. Still, he said he would consider using DALL-E 2 for tasks such as providing editing guidelines and conveying ideas to graphic artists. “For stock photos without people and illustrations without specific brand guidelines, DALL·E 2 could, to my non-expert eye, reasonably replace the ‘old way’ at this point,” continued Silberstein.
Editors at Cosmopolitan came to a similar conclusion when they teamed up with digital artist Karen X. Cheng to create a cover for the magazine using DALL-E 2. To get to the final cover, there was a very specific question from Cheng, which the editors said was the illustrative of the limitation of the DALL-E 2 as an art generator.
But the madness of AI sometimes works – as a feature, rather than a bug. For his Draw Ketchup campaign, Heinz had DALL-E 2 generate a series of images of ketchup bottles using natural language terms such as ‘ketchup’, ‘ketchup art’, ‘fuzzy ketchup’, ‘ketchup in space’ and ‘ketchup renaissance’. The company invited fans to send their own prompts, which Heinz has curated and shared through his social channels.
Heinz bottles as “imagined” by DALL-E 2, part of Heinz’s recent ad campaign.
“With AI imagery dominating the news and social feeds, we saw a natural opportunity to expand our ‘Draw Ketchup’ campaign; rooted in the understanding that Heinz is synonymous with the word ketchup — to test this theory in the AI space,” said Jacqueline Chao, senior brand manager for Heinz, in a press release.
Obviously, DALL-E 2-driven campaigns can work when AI is the subject. But several DALL-E 2 business users say they’ve used the system to generate assets that don’t bear the telltale signs of AI limitations.
Jacob Martin, a software engineer, used DALL-E 2 to create a logo for OctoSQL, an open source project he is developing. For about $30 – about the cost of logo design services on Fiverr — Martin was given a cartoon image of an octopus that looks human-illustrated to the naked eye.
“The end result isn’t ideal, but I’m very happy with it,” Martin wrote in a blog post. “As far as DALL-E 2 goes, I think it’s still very much in a ”first iteration’ phase for most parts and purposes at this point – the main exception being pencil sketches; those are stunningly good… I think the real breakthrough will come when DALL-E 2 gets 10x-100x cheaper and faster.”
The OctoSQL logo generated after several attempts with DALL-E 2.
A DALL-E 2 user, Don McKenzie, head of design at dev startup Deephaven, took the idea one step further. He tested the application of the system to generate thumbnails on the company’s blog, motivated by the idea that posts with images get much more engagement than posts without.
“As a small team of mostly engineers, we don’t have the time or budget to do custom artwork for all of our blog posts,” McKenzie wrote in a blog post. “Our approach until now has been to spend 10 minutes scrolling through tangentially related but ultimately ill-fitting images from stock photo sites, downloading something not terrible, putting it in the foreground and pressing publish.”
After spending a weekend and $45 in credits, McKenzie says he was able to replace about 100 blog posts with DALL-E 2 generated images. It took some fiddling with the directions to get the best results, but McKenzie says it was well worth the effort.
“On average, I’d say it took a few minutes and about four to five prompts per blog post to get something I was happy with,” he wrote. “We spent more money and time on stock photos each month, with a worse result.”
For companies that don’t have time to brainstorm, there is already a startup trying to commercialize the asset-generating capabilities of the DALL-E 2. Unstock.ai, built on top of DALL-E 2, promises “high-quality graphics and artwork on demand” – currently free. Customers enter a prompt (e.g., “Top view of three goldfish in a bowl”), then choose a preferred style (vector art, photorealistic, pencil) to create images that can be cropped and resized.
Unstock.ai essentially automates prompt engineering, a concept in AI that seeks to embed a job description in text. The idea is to give an AI system detailed instructions so that it reliably does what it’s asked of it; In general, the results for a prompt such as ‘Movie still of a woman drinking coffee, walking to work, telephoto’ will be much more consistent than ‘A woman walking’.
It is probably a harbinger of future applications. When contacted for comment, OpenAI declined to share figures on business users of DALL-E 2. But anecdotally, the question seems to be there. Unofficial fixes for DALL-E 2’s lack of API have been popping up all over the web, brought together by developers eager to embed the system into apps, services, websites, and even video games.