AI image generation remains controversial for several reasons, and at the top of the list is the issue of copyright. Several text-to-image generators were trained using images scraped from the web without permission, and some are able to replicate, or at least attempt to recreate, the style of specific artists, who receive no credit or royalties.
Tess promises to be different. It’s the first AI image generator I’m aware of that’s being billed as a direct collaboration with artists. It specifically aims to copy and replicate the style of its collaborating artists and pays them royalties when their style is used.
Tess came out of beta this week, and it’s being billed as the “first ethical AI image generator”. If that phrase sounds familiar, that’s because it was used a lot on the launch of Adobe Firefly last year. However, there’s a key difference in Tess’s approach.
Adobe argues that Firefly is more ethical in comparison with the likes of Midjourney, DALL-E and Stable Diffusion because it was trained on licensed images from Adobe Stock, where contributors receive at least some form of compensation. It also includes content credentials clarifying that images were produced by AI. But just how ethical it is has been questioned. Some Adobe Stock contributors complained that they weren’t able to opt out, and it’s been pointed out that Adobe Stock itself contains AI-generated images, including images made using Midjourney.
What makes Tess different is that it’s built to specifically copy the work of collaborating artists through a series of dedicated individual models. The artists are named, they license their work, and they receive royalties when their style is used. On launch, there are 15 artists on board, including David O’Meara and Jose Elgueta.
Tess has been developed by San Francisco-based Kapwing, an online video editor founded by Julia Enthoven and Eric Lu, who previously worked on Google Image Search. It currently offers 22 different models, each fine tuned using 10 to 20 works created by in a particular artist in consistent style. Users can choose the style they like and use text prompts to generate the images they want. Tess then splits subscription revenue 50/50 with participating artists based on the number of generations that use their model or models. Tess says it’s paid out over $15,000 in royalties so far.
As for the customers, the image generator is aimed at creators, journalists, media entrepreneurs and small business marketers. Customers pay a base subscription is $20/month and get access to embedded tools for editing images and generating prompt ideas.
Daily design news, reviews, how-tos and more, as picked by the editors.
Tess says it’s also the first AI image generator that gives artists a choice on how they want to use Generative AI. Contributors can set up a private model to explore the technology, or they can choose to list it in the public marketplace to earn royalties. Tess provides attribution guidelines that aim to ensure artists get credit and visibility when images generated using their models are shared, and clients can also contact artists directly through the info on their profile.
So is it too good to be true? Well, I noted already that each model is trained on just 10 to 20 works by the collaborating artist. The underlying model behind the whole thing is Stable Diffusion, the well-known open-source AI-image generator whose developer, Stability AI, is being sued by artists for alleged copyright infringement.
Enthoven says she sees no contradiction. She says Tess has protections in place to prevent its models from producing images in the style of an unlicensed artist and that the additional training on specific artists’ images means that outputs are significantly transformed from what Stable Diffusion would produce alone.
That might not be enough to convince some that a truly ethical AI image generator can ever exist. That said, Tess does appear to offer a way for artists to at least get something out of it, both financially and in terms of exposure. The debate for those considering joining will be whether it opens up a new avenue for their work or risks losing customers who could choose to use their AI model instead of commissioning them for original work.
Andrew McGuire, an artist and muralist whose work was used to train a model called All of Us, said: “Tess offers a great opportunity to utilize my work in ways I could not fathom and lets me participate in the AI conversation, not be left on the sidelines,” said
The artist Ger Jorge said: “I think that AI is here to stay, and it is counterproductive to oppose something that will remain a part of our future, and that its greatest virtue is to somehow democratize art and its creation. Just as the person who tightened screws was affected by the arrival of an automatic tool, surely that person had the power to become the one who fixed that tool when it broke down. Today we artists can educate this intelligence so that anyone with artistic desires can carry it forward and Tess, in addition to carrying this project forward, which is no small feat, fulfills the most important part, which is to give credit and effective participation to the artist.”