How SMBs Can Benefit from Text-to-Image AI

Getting the chance to try DALL-E 2, the new AI system from OpenAI who can create realistic images of natural language was quite extraordinary.

There’s no doubt that the system is still in its infancy, but it’s clear that the technology is moving fast and we’re already starting to see improved text-to-image models. Google Brains Imagethat can generate photo-realistic images of a scene with a textual description and Metas To make-A–Sceneallowing users to draw a free-form digital sketch at a text prompt are both promising examples.

This technology isn’t just for big tech companies either. There are a number of ways that small and medium-sized businesses can take advantage of text-to-image technology today.

On average, it is recommended that small businesses spend 7-8% of their gross revenues on marketing. And yet many spend only 3-5%. Tools like DALL-E 2 allow entrepreneurs to stand above their weight, even if they don’t have the luxury of hiring talent to produce customizable, branded images.

As the saying goes, a picture is worth a thousand words. As DALL-E 2 and other text-to-image systems become more widely available, it makes sense that marketers would start using them for their campaigns.

Take, for example, a brand that sells fresh seafood. While it makes sense to include tempting images in the emails, there’s no real advantage to doing one’s own photo shoots over using stock photos. Subscriptions to stock photo catalogs for commercial use can easily cost hundreds of dollars a month, which is hard to justify for smaller brands.

Enter DALL-E 2. The model allows marketers to combine the creativity of hiring an in-house graphic designer with potential price savings. For example, this is what DALL-E 2 produces when he asks for “a photo of a mouth-watering salmon with lemon slices.”

These images are indeed mouth-watering, but what happens if the supply of salmon unexpectedly drops and our supply shifts from salmon to cod this week? For small businesses that need to change in the blink of an eye, the ability to generate new images in 30 seconds is invaluable.

Graphic creation by DALL-E 2

DALL-E 2 has entered the popular consciousness not for its ability to faithfully mirror the world, but for its ability to create beautifully stylized images.

Consider this watercolor of a panda wearing a hat. The image is playful, well-structured and doesn’t appear in any stock photo catalog. That’s why DALL-E 2 is so exciting: it opens up new opportunities, especially for small businesses.

The success of tools like DALL-E 2 could be discouraging for designers concerned about AI automating their roles, especially as some experts predict 99.9% of online content will be generated by AI by 2030. But we believe that DALL-E 2 will not replace jobs; it will instead become part of a marketer’s toolkit and skills.

The output quality of the DALL-E 2 varies greatly depending on the prompts entered into it. Generating a good prompt is a very creative process, similar to writing a copy yourself. If the future includes marketers in hybrid roles using these tools for inspiration, then this will increasingly become a domain skill.

DALL-E 2: Paint in

One of the most exciting features the DALL-E 2 offers is automatic painting. Think of it as a magic wand that marketers can wave to change images however they want.

This is best seen with an example. Here, DALL-E 2 was given the first image and told to insert a corgi in a specific location. The model understood the context of the image well enough to understand that it was asked to depict the dog in a painting and could compose in a style to match.

This fascinating feature has a lot of value, especially for brands looking to capture their products in different locations or unique scenarios. Imagine putting your product on a backdrop you can’t visit, or seeing what George Washington would have looked like sipping a Coca-Cola.

DALL-E 2 can remarkably change the game for marketers where they are limited more by their creativity than by time and budget.

Limits

While there is reason to be excited about the future, current technology still has limitations. It struggles to generate photo-realistic people and often fails to generate coherent text. In addition, there are some significant criticisms of DALL-E 2 on the issue of bias. For example, when he was tasked with generating “portrait of a smart person,” DALL-E Mini — a cheaper, open source implementation of DALL-E 2 — generated nine photos of white men in formal attire.

OpenAI recently released an update aimed at reducing some of these biases, but it remains to be seen whether they really get resolved. In the meantime, marketers using these tools will need to be careful to avoid unintended bias.

Shut down

So do we want DALL-E 2, in its current form, as a tool in our marketing toolbox? Absolutely, and for many purposes: images that can be used as generated, to provide clues for further editing; help brainstorm; to convey ideas to graphic artists; perform efficient A/B testing on creatives; and to create fantastic, eye-catching images that can only be created by artists today. Despite the limitations, there are also opportunities.

Over the past six months, we’ve seen a growing interest in groundbreaking text-to-image technology. As technology continues to evolve rapidly, better, improved models will emerge and use cases for brands will only increase.

Robert Huselid and Tom Dinitz are data scientists at Klaviyoa unified customer platform for email, SMS and more, empowering online brands to own their data and grow on their own terms.