Bringing back Dr. Seuss
Can AI continue the legacy of our most beloved authors and artists?
Today I experienced something both amazing and terrifying.
But first, let’s talk about Dr. Seuss. If there was one author who shaped my childhood imagination, it would be him. His unique style of writing, with its signature rhyming and use of made-up words, was unlike anything else I had ever read. Dr. Seuss was what made reading fun for me, and to this day, I use his books to instill a love of reading in my own children. But Dr. Seuss was not only a great writer, he was also a brilliant artist. His illustrations were just as iconic as his words, and together they create a world that is both fantastical and inviting.
Sadly, Dr. Seuss passed away in 1991. As many of you already know, I am very excited about the emerging space of AI art, and this has me thinking, what if we could apply AI to bring back the art of Dr. Seuss, 31 years after his death?
Well, we are not there yet, but I tried something today that blew my mind. I scanned twenty-five of Dr. Seuss’ illustrations from my children’s book collection and fed them into a new AI algorithm called DreamBooth. DreamBooth is a fine-tuning method for Stable Diffusion (or other image synthesis AIs), invented by Google Research, that is capable of some pretty wild things. I had used DreamBooth and Stable Diffusion to produce some impressive results before, but I was very doubtful that it would be able to accurately emulate the signature style of Dr. Seuss.
I was wrong.
The results were astounding. The new Seuss-AI model was able to take the style of Dr. Seuss and use it to generate new illustrations that look like they could have been done by the man himself. I have never seen anything like it. The colors, the shapes, the lines, all impressively captured the Seuss-ness of it all.
Admittedly, as you can see in the examples at the bottom of this article, it’s far from perfect. Even the best cherry-picked results don’t come close to the creative genius of Dr. Seuss. Some images still have obvious defects such as missing eyes or twisted limbs. But considering that I used only a handful of images to “teach” the AI the style of Dr. Seuss, and considering that the entire training process took under an hour, I am totally mind blown by how it is able to capture the style and nuances so well (as a side note, the base, non fine-tuned Stable Diffusion model contains some prior knowledge about Dr. Seuss, but without the DreamBooth fine-tuning process, this model produces very poor results).
I am still in shock. I cannot believe what AI is capable of today. I’m not sure what the future holds, but it’s thrilling to think that a more advanced version of this technology might one day be able to bring back the art of beloved authors and illustrators like Dr. Seuss. Is that a good or bad thing ? Today, great artists and writers are immortalized by the work they have created during their lifetime. But what happens when AI can continue their legacy ? We might be living in a world where the works of Dr. Seuss never end.
And that, my friends, is both amazing and terrifying.
Thank you for reading! If you liked this, follow me on Twitter:
