StyleDrop is an AI that does not seek to push the limits of creativity. His instruction is to stay as close as possible to his model. And it performs its task successfully.
Playing with generative AIs is one thing, finding a professional use for these new tools is another. When it comes to generating images, AIs have seemingly limitless creativity, delivering amazing results in response to quick.
It is this effervescence that we are looking for first. But if creativity seeks the unique and the new, the professional world often needs elements that look alike. A graphic charter, typography, flyer, web page, artistic direction or signage are all elements that require visual consistency. This link is currently one of the weak points of AIs: even with a similar prompt, you can change universes.
It is to meet this need for semantic and visual consistency that the StyleDrop AI was created by Google engineers.
StyleDrop, an AI that knows how to duplicate a style (and stick to it)
StyleDrop can be summed up in a very simple way: from a visual (an image) or textual prompt, the tool knows how to replicate the style by understanding its essence and applying it to other requests. This is how he manages to be consistent in his creativity.
For example, StyleDrop is able to create wooden structures in the same style, with a unique background. We notice that there is, at first sight, no difference between the approach to generate an image or another. They could pass for creations of the same artist.
Even more impressive: from a simple photo of a watercolor drawing, the AI created an extremely consistent typography. She could have hallucinated very different colors and shapes, but chose to keep the essentials: simple lines, which imitate water-based paint. We immediately see the use of this function to create complete and creative typographies, directly usable.
Equally important: the generation of logos and pictograms. This task that Midjourney struggles with is pretty well handled by StyleDrop. By taking black pictograms as a reference, with several styles, the AI generated a coherent sequence that could appear on a web page. We could see the limit if we dug a little reflection: in this selection which seems coherent, two styles actually mingle. The first (wolf, car, eye, fox) is characterized by black lines and empty spaces that materialize the forms. The others are solid black shapes that materialize the images.
The abundance of concepts related to AI can sometimes make us lose sight of the essential: what are they for? Can they help us on a daily basis beyond the first hours of discovery? Many projects answer these questions in the negative. StyleDrop, on the contrary, could well represent the future of a category of generative AIs, which meet a real need.
In any case, we may use it to improve our Artificial newsletter to which you should subscribe if you are passionate about this universe!
Subscribe for free to Artificials, our AI newsletter, designed by AIs, verified by Numerama!