The problem that started Tess

This month, we’re introducing Tess, an AI generator that is trained on the visual style of an artist, with their permission. In this article, I want to share the problem that Tess is trying to solve. 

Artists are getting screwed

We’ve been paying close attention to the development of new generative AI technologies. 6 years ago, my cofounder and I started an online image & video editor that helps creative teams tell stories. Dall-e - the Image generator built by OpenAI - is popular within Kapwing, and we’ve seen new video generator demos from Sora and Runway. 

Most of these companies, however, have been built without permission from or partnership with artists. We’ve seen the lawsuits as artists and creators have protested about their work being used in proprietary models with closed training sets. 

In the last few months, we have had conversations with dozens of artists about this problem. Artists are frustrated as some commissions have dried up where people are using generated art instead. They’re also angry, indignant, and scared, unsure how this new technology will impact them and their careers. AI companies argue that their training data qualifies as fair use, but it doesn’t feel very fair to artists.

The Society of Illustrators has shunned generative AI completely, as have many illustration agencies. This has limited the ways that the industry has benefited from generative AI, as artists are indigent about using the tools to brainstorm ideas, transform sketches into designs and characters in their style, create promotional materials, and make low-fi mockups for clients. 

And we understand why. It’s incredibly frustrating and scary to see a technology that is faster and more prolific but dumb, low-quality, and uncontrollable displace some aspect of your work. Experienced illustrators have often taken years to master and refine their craft. 

The status quo reminds us of the music industry in 1999, when everyone was stealing songs and ripping them off the internet. Back then, we needed a new business model that empowered the artist and the consumer alongside the rise of the internet. Along came iTunes, and eventually Spotify.

Generative AI is too general to be beautiful

Publishers, meanwhile, are hesitant due to the ongoing legal questions, but also because generative AI is hard to use. It’s too literal and photorealistic, requiring specific knowledge to enter a prompt that’s detailed enough to craft what you want. DALL-E users have few tools for altering a generated image. If you want to change the aspect ratio or re-generate one element within the image, you often have to start from scratch with a new prompt.

At Kapwing (our first business), we ran into this when we tried generative AI for the illustration work for our own website. We have an in-house illustrator, Emily, who is extremely talented. She was bogged down with daily fast-turn-around requests from our content marketing team for blog post and youtube graphics, and she wanted more headspace to focus on new skills and more complex projects.

Our content lead guided his team to start using DALL-E, since we already had a company license. However, we found the generated outputs robotic, almost like out of a video game or sci fi movie. How could we preserve the illustrated style and brand of our Resources Library without needing Em’s time and attention on every blog post and newsletter?

So, we built a custom AI model trained on Emily’s work (with her permission) that enabled our content marketing team to generate their own graphics. It’s a much better experience than using Dalle as it has features custom-built for our writers, like in-painting and the ability to change the output aspect ratio to an OG image size. 

That was the first Tess model.

Introducing Tess

Introducing Tess, a platform for editorial and marketing teams to generate illustrated work. Artists have a custom model trained on their visual style, with their permission, and get paid anytime that a Tess customer generates images using their model. This passive income (coming in from Tess customers around the world) can make up for and perhaps surpass the work that was lost from the high end of the market. 

We believe that clients will prefer Tess over other AI image generators because they want to foster a healthy ecosystem and want full copyright protections with properly-licensed work. They may also find that the works generated from Tess are much more beautiful than the stock photos they’ve used in the past. Instead of starting from scratch, they explore a marketplace of visual styles to find the right look for the right piece.

We envision a world where artists have more control of and knowledge about generative AI. Tess Artists can access their own model for free, and Tess will provide resources for artists on mastering generative AI as a tool in their practice. We also plan to give them visibility and control in how their model is used.

A couple of weeks ago, I met with an accomplished artist called Owen who has in the past done a lot of freelance illustration work for independent musicians in the past. He told me that 75% of his commission workload for musicians has dried up in the last year. He can't be exactly sure why, but when he looks at the new album art of people who used to be his clients, he thinks they used generative AI instead of hiring him. Those products were likely trained on his work with no permission from or royalty to him.

Owen (not his real name) is joining Tess because he wants to make up for that lost income from those musicians’ commissions, and potentially reach the hundreds of small bands out there who couldn’t afford to work with him before. That would be a Tess success story.

Join us

• If you’re an illustrator who is angry about AI companies stealing your work and displacing your commissions
• Or an artist who is curious about generative AI and wants to collaborate with it to scale your own career
• Or a small business owner who wants to support illustrators but can’t afford personalized commissions
• Or a writer who is tired of boring, redundant stock photography 
• Or a publisher who wants to use generative technologies with proper copyrights

Tess is for you

We have more than 20 Tess models and are currently in beta. We're looking to partner with agencies, media companies, and professional organizations to get the word out about Tess. Get involved by emailing us at hello@tess.design.