Tess LogoTESS
Back to all posts

How we solve AI image generator copyright infringement

We started Tess, a properly-licensed and commercially-safe image generator for editorial and marketing teams. Here's how we addressed the issue of copyright infringement and created a pro-artist business model.
How we solve AI image generator copyright infringement
Image by Foreign Nolstagia on Tess

We’re the creators of Tess, the first properly-licensed AI image generator. Tess clarifies the ownership of AI art through a licensing agreement and a model trained in a single artist’s visual style. The artist owns their style and gets paid each time it's used. 

The problem

Does the world need another AI image generator? In a word, yes.

No one feels good about using current AI image generators because they cut out the artist entirely. Although most small creators can’t afford to commission illustrators, they don’t embrace generative AI given the anger and frustration from the art community.

In a recent episode of Hard Fork, for example, tech journalist Casey Newton commented that he recently stopped using AI generated images in his newsletter, Platformer, because he “got a lot of feedback from readers that they hated it” and “felt like [he] was stealing money from artists.”

Many companies - including local newspapers under the Hearst umbrella - have prohibited their publications from using AI image generators due to ongoing legal concerns. OpenAI, Midjourney, StabilityAI, Microsoft, Meta, and Google are all facing large lawsuits related to their image generators outputting infringing material. Journalists we’ve talked with say that they want to produce more original art but are barred from the tools by their legal team. 

In addition, the existing AI tools are hard to use. Unless you’re a trained prompt engineer, most Midjourney and Dall-E results look robotic or straight out of a sci-fi movie.

AI-generated images can look, well, AI-generated. (from an AI Substack)

With increasing restrictions on image generator prompts, it has become harder - not easier - to produce good-looking art. Serial creators and brands find that it can be difficult to create images with a consistent visual style.

Screenshot from ChatGPT

So, we’ve set out to build an alternative that people feel good about using, that empowers rather than cuts out artists, and that enables people to generate beautiful art in a consistent style. 

Image by Foreign Nolstagia on Tess

What is Tess?

Tess is an AI image generator that serves editorial teams, media companies, and creators. On Tess, each AI model is trained in a single visual style, with permission from the artist, and the artist gets paid each time someone downloads an image. 

We buy the rights to generated works from the artist through a transparent contributor agreement. The artist works with Tess voluntarily (and exclusively) and defines what 10-20 works will be used to train the model. Because of this, we can sell an editorial and commercial license to the generated artwork.

Image by Foreign Nolstagia on Tess

Importantly, Tess does not output images in a style it does not have permission to reproduce (unless that style is in the creative commons). We protect artists who choose not to license their work to us by preventing users from generating in their style. While Dall-e and other image generators prevent users from generating in the style of a living artist and block prompts of copyrighted characters, Tess enables this powerful use case by helping living artists own, share, leverage, and profit from their own AI generator.

As a result, creators and publishers can generate images on Tess without fear of copyright infringement. 

A Healthier Ecosystem

When someone uses Tess, they’re supporting the artist who authored the style. Because it reduces the cost and increases the speed of producing art, Tess opens up a large segment of the market that before could not afford custom commissions, creating a new passive income stream for artists to make up for displaced work. Publishers can connect with the artists they discover through the Tess marketplace, and artists have more visibility and control over who is using their model and how.

“What Tess does is great because it works directly with the artists, so that there is still a human aspect of the artwork,” Andrea Sanchez, a Tess artist, explains when asked why she joined the marketplace.

Tess was built in partnership with artists, ensuring proper attribution and royalty payments, and we have about 25 accomplished artists onboard. 

We’re also in talks with several of the country’s biggest news organizations who are seeking an alternative to the legally-fraught AI image generators on the market. Small creators - like musicians making band posters or bloggers making cover images – can feel good about using AI as their usage payments will fund artists who have consented to put their style on Tess.

Image by Foreign Nolstagia on Tess

Tess is designed to accelerate the design and illustration process, helping the artist to scale their own creative capacity. Artists get access to their own model for free and leverage built-in ideation tools for prompting and editing. With features for brainstorming, transforming sketches into images, and making mockups in their visual style, design teams can use Tess as an accelerant rather than a replacement for human creativity. 

The Tess licensing model can be applied to other media outside of image generation. In the future, we plan to expand Tess to voice, music, animations, videos, and graphics in the future.

Why we believe in the Tess future  

After dozens of conversations, we believe that Tess is a better solution for all involved than the current generative AI applications. It enables artists to leverage the technology themselves and to benefit when their style is used in the wild. Tess is the first place where an artist can “own” their visual style. They can use it to free up their time and focus on more complex projects.

Image by Foreign Nolstagia on Tess

We hope that creators, individuals, and small businesses will find Tess much easier to use and feel better about how they contribute to the artistic ecosystem. For media companies, we envision that Tess will enrich the relationship between artists and publishers, enabling them to collaborate more effectively, rather than blocking the human out. 

Not everyone agrees that Tess is good for the world. For artists who don’t want to be on Tess, we respect your choice; users won’t be able to generate in your visual style on our platform. 

Thanks for reading! Let us know your thoughts and join the waitlist at tess.design.

Julia Enthoven

Cofounder and CEO, Tess

How is this different from other “ethical” AI image generators? 

Some image generators - like those by Getty Images and Adobe - claim to be more ethical because they’ve been trained only on licensed work or images in the creative commons. This can reduce the quality of outputs significantly as the model has learned from far less data, but it also means that the attribution and inputs are untraceable. 

As a result, no other platform - to our knowledge - pays usage royalties to illustrators when their visual style is used or enables users to generate in one consistent style. (Adobe Firefly also allows users to generate based on a reference image that they may or may not have the rights to use). 

In contrast, Tess images are defined and transformed by a signal visual style from one artist licensed specifically for that purpose, making the credit and royalty model more clear for both sides. 

How do you train a model on such few images?

Tess uses a custom pipeline loosely based on SDXL to generate images. Each model is adapted and fine-tuned on the artist’s visual style. Unlike AI generators by Shutterstock, Tess benefits from a large foundational model, but significantly reduces the risk of copyright infringement because the visual style outweighs (and significantly transforms) the underlying data.

Tess gets better and better as the underlying models improve, which means that the artist stands to build more passive income through usage royalties over time.

Does Tess dilute the value of the artist’s work?

We believe that new tools increase the value of original ideas rather than diluting them. As Tess grows, the impact of the artist's style grows too. More people can leverage it to create original art, although the artist will still own the nuance of their visual concepts and innovative ideas.

Given that artists give up some control when joining Tess, we’ve taken steps for them to preserve the value of their personal brand. The artist can give their Tess model a generic name, a pseudonym, or their real name for attribution. Also, Tess images are watermarked with a C2PA-protocol metadata to help distinguish them from original works. Read more on tess.design/artists

Did you work with a lawyer on Tess?

Yes. We have worked extensively with a senior lawyer through the law firm Fenwick to draft and iterate on a novel Contributor Agreement and licensing agreement. 

Who else has contributed to Tess?

Thanks to the following people who have met with us about Tess: Jillian D’Onfro, Steven Heller, Jamie Jiang, Katherine Schwab, Byron Walker, PJ Loughran, Yiren Lu, Sarah Roberts, Anjana Susarla, Lisa Iaboni, Chris Lyons, and more than 40 different professional illustrators.