Controversy continues to surround the world of artificial intelligence-generated images
The controversy continues to surround the world of artificial intelligence-generated images, and even though this week at the World Economic Forum AI-generated images used in elections have become a cause for concern, startups continue to develop new AI tools for creators.
The latest one is Recraft, an AI graphics generator aimed at professionals, which has raised a $12 million Series A led by Khosla Ventures in Silicon Valley, along with former GitHub CEO Nat Friedman. RTP Global, Abstract VC, Basis Set Ventures, Elad Gil, and various other angel investors also participated.
To be fair, there are now many generative AI design tools like Jasper, Adobe Sensei, Let's Enhance, and many others. However, Recraft claims to be one of the first to be a "fundamental" tool, in the sense that it is building its Fundamental Model - a pre-trained deep learning algorithm - to generate consistent design elements, such as icons and images, that can be customized and used within a brand's style guidelines. It also claims to have collected over 300,000 of them since its launch eight months ago.
But this isn't designed as a tool to generate funny images of cowboy dogs riding horses, or the like. In addition to being able to produce these "raster images," it can also generate infinitely scalable vector images used in the professional realms of graphic design, unlike platforms where image quality is often limited.
Perhaps this is to be expected, as the founder, Anna Veronika Dorogush, is not exactly a non-technical founder. She created CatBoost, an open-source high-performance library for boosting on decision trees, for example. A former head of machine learning systems at Yandex's search engine in Moscow, Dorogush and her team of 13 people are now based in London.
She told me on a call that the fundraising will fuel her attempts to build their own Fundamental Model: "You need to provide a lot of controls over the outputs to the users... over the style, so you can get consistent images, and control over things like brand colors or level of detail, and also the ability to iterate on the resulting image."
"But if the initial model can't generate a pool player or a classical dancer, this means that just using fine-tuning is not enough to provide high-quality generation, for now. So we have to build our model, and we are working on this right now," she added.
She says Recraft is very geared towards professionals: "What's unique in what we're doing is that we don't just provide image generation. We also provide style control: the ability to create your own style and then generate images in your style... This is important if you want to create a brand and grow it, create marketing materials, create consistent ads."
I asked her about the likelihood of them being sued by an artist whose style was used on the Recraft platform.
"In our terms, we state that if an artist uploads something into the system as a style reference - like the image for which they will generate new images - we don't use those to train our model," Dorogush says.
"We don't use those to train our model."
She added that Recraft effectively adds generic styles to the images uploaded by artists: "You provide the style to the model. So you have a style reference. And then you generate images using this style reference. So it's the action of the user providing the style."
Dorogush also argues that, rather than replacing graphic designers, tools like Recraft will simply make graphic design more accessible in areas where uninspiring stock images may have been used in the past: "A year ago, if you were writing a children's book, you would get stock images. Now, the same people are able to have much more beautiful illustrations and images in the same book."
Of course, Recraft is not the only startup playing in this space, and last year I covered how Berlin-based Kittl raised a $10.8 million Series A (€10.8 million) for its design platform that allows users to turn ideas into graphic products to create high-quality designs quickly.
This week, CEO Nicolas Heymann told me on a call that products like Recraft are part of an increasingly sophisticated set of tools providing alternatives for designers: "I think if Recraft can successfully produce design assets - in a similar way that image generation models can be a good replacement today for stock photo libraries - then they could certainly take [a part of the market] of those icon libraries."
Dorogush says there is room for more than one player right now and that Kittl has a different use case from Recraft: "We have unique technology for building consistent images in style, and that's where our main focus is. We are focusing on designers and marketers who need style consistency, whereas Kittl focuses on merchandise, and I have great respect for them. They are a great company."