First published on October 27, 2022, it’s one of our favorite pieces of the year.
Playing around with the AI art tool Stable Diffusion. Automatic1111 Web UI version (opens in new tab) First launch. I’m not very command line savvy, so having a simple mouseable interface is pretty rewarding for me. And for men without visual art bones in their bodies, it’s a fun plaything. I thought of The Hitchhiker’s Guide to the Galaxy, Boris Johnson’s painting of Monet sitting on a toilet in the middle of a pond, my dear Donald reading his PC format with his playing cards.
But nothing inspired me more than hitting the Nvidia RTX 4090 (opens in new tab) For eight and a half hours straight, I trained to paint like my great-uncle Hermann.
You don’t know the name Hermann Kahn. Also, I would be incredibly surprised if I actually recognized him by a widely known name. Aharon Kahana (opens in new tab)To be honest, I didn’t even know him. Sadly he died long before I was born.
But growing up, I’ve heard so many stories about Uncle Hermann, both from my mother and my late grandmother, that I feel like I know him. At least part of him anyway.
Kahana’s art style is unique and a hallmark of my childhood.
The family bond has grown stronger since we traveled to Tel Aviv just before the birth of our 3-year-old son. It was where my grandmother, Inge, and her great-grandmother, Rosa Kahn, fled from pre-Krystal Nacht Germany in the mid-1930s. and where Hermann Kahn settled after meeting his wife while studying art in Berlin.
I walked the streets they walked, past the apartment where my grandmother grew up, traveled the road Haifa Rosa took every morning, and visited Hermann’s house in Ramat Gan.
image 1 of 3
The house he shared with his wife Midée became a museum for his art. It was closed when I visited, apparently it had been closed for a while, but has since reopened and is now hosting exhibitions again.
Kahana’s art style is unique and a hallmark of my childhood. I was surrounded by his pottery and paintings in both early and late styles at his parents and grandparents’ homes. Even as a child, I was drawn to them.There are certain vases that I never could No It’s considered the Spaceship Enterprise thanks to its Trek-like saucer section.
The completely abstract geometric imagery I always envisioned would look like a loving couple adorning our chimney chest, an image of Parisian roofs and a storm painted in thick oil paint. A beach scene ran up our steps.
Inevitably, however, the early 20th-century German-Israeli painter and ceramist is not one of the artists on Stable Diffusion’s list. And while I’ve experimented with detailed prompts and fiddled with X/Y plots to find levers to pull to get an approximation of the abstraction he’s created, I can’t really get there. There was no.
The Stable Diffusion checkpoint file is missing the required reference points. But there are ways to encourage AI to understand different related images and build from them concretely. They’re called embeddings, and people used them to train tools to recognize their own faces. That way you can include yourself in all the wild hairy AI-drawn fantasies you want.
However, I wanted to train the AI to recognize and understand the art of Aharon Kahana as much as possible with a relatively simple AI. It’s a surprisingly powerful tool, especially given the caveat in the embedded description that “features are very raw, use at your own risk”.But the latest in his Web UI app on Github Thanks to the release, you can do it all from your browser.
image 1 of 3
It requires Stable Diffusion, so Python must already be running on the machine, but when I put together a folder of images with a specific name, the GPU is thrashed to 100% load, and the load drops to 50%. increase. CPU, time to create a reference point that Stable Diffusion can use when prompted for the exact name of the embedding.
I wanted to train the AI to recognize and understand Aharon Kahana’s art as much as possible with a relatively simple AI.
It sounds relatively simple, but it certainly took some trial and error. In particular, after downloading his 70 or so images of my great-uncle’s work from various auction sites around the world, I actually had to label them with vague and detailed labels for the training to be effective. I noticed
It took me quite a while to figure out the medium and subject of each piece I downloaded, and to manually rename each file. And it’s not always so easy when you’re dealing with very abstract images.
Then I pointed the RTX 4090 and my Core i9 10900K at the relevant folder, created an embedded wrapper, and let it sit for over eight and a half hours, agreeing to what I fed. All of his 16,432 cores and a decent chunk of 24GB of memory on the new Nvidia card, and half of my 10th Gen Core i9 were used for this task.
I don’t pretend to be smart enough to truly understand what I’ve put on the world’s most powerful consumer GPUs, but when I looked it up at night, I discovered that I was getting the input image. understood. Create your own approximation.
image 1 of 2
It was like my PC spent the night learning from Hellmann and doodling down homages to his style, trying to figure out how to do it without the artist’s help.
By morning the embedding is complete, I fire up the web UI again (it now lists one text-reversed embedding), paste the text “by aharon_kahana” at the end of the prompt, and see what the AI does overnight. I was able to check if I had learned
And it was amazing. My computer was churning out homages to my great-uncle. I’m a complete novice when it comes to prompt mystical art, but even my basic request provided images that evoke the artist’s memory.
What they lacked in pure soul and understanding of what they were really doing, they made up for in quirky digital creativity and GPU-backed efforts. Indeed, it was all recognizably and closely related to his art his style.
image 1 of Five
I know many contemporary artists are against the development of AI art. I know you are frustrated by the abundance of pictures of fantasy women created by people with no artistic talent. I can’t help but think he would have embraced this new tool.
That’s the tool. I was impressed with how close Stable Diffusion came to recreating his art his style, but all Stable Diffusion can really do is recreate it. I’m not going to evolve the style by itself. Human artists need to evolve art further. And it needs detailed human input to give it enough subject matter to build on.
It’s not a replacement for artists, it’s just a tool to be built into the arsenal of artists interested in taking technology to new and interesting places, much like high-definition SLRs and Photoshop came to landscape painters. .
So at its current level, AI art feels like a starting point, rather than something that can truly create a finished product. That said, you probably won’t stop filling your PC with a myriad of colorful, endlessly abstract images. All inspired by some of my family members.
