Latent Space Netsuke
July 28, 2022 9:33 AM Subscribe
Latent Space Netsuke
I know it's starting to get a little cliché with the AI generated image projects, but I'm pretty pleased with this one: A series of photo-realistic works of imagined "netsuke" sculptures, using DALL-E. I've created over a hundred, which I'm posting online (in addition to the Instagram account, there's a twitter account for those who prefer it), and collected 80 into a print book for people who are into that sort of thing.
Here's the intro from the book:
I’ve always been fascinated by figurines. As a kid, it was chess pieces and gaming figures, then Gothic boxwood miniatures from the 16th century. Later, I discovered the many exquisite forms of netsuke, the Japanese sculptural objects made famous during Japan's Edo period. Netsuke originally had a practical purpose: They were used to attach a box or pouch onto clothes that lacked pockets, but their aesthetic qualities, both in the diversity of subject matter and the quality of craftsmanship, were always evident. Netsuke artisans experimented for over two centuries using a variety of materials (although primarily wood and ivory) to create these sublime works of art.
The images in this book were created using a new computer program called DALL•E, developed by the company OpenAI. The program takes a string of text (known as a “prompt”), such as “A cow standing in a field during a sunset” and returns an image that is the best approximation it can find to match that text. How does it work? Essentially, the program was “trained” on millions of pairs of images with text descriptions to “learn” the connections between elements of images and certain words. This technique resulted in a digital model of the billions of inter-connections between these elements. These connections are sometimes referred to as a “latent space”-- a space that contains a near infinite number of potential images.
In my early experiments with DALL•E, I stumbled upon a prompt that could, with some regularity, produce photo-realistic images of something approximating a netsuke. These are not actual netsukes, of course. Although they may look "real", these are pure works of digital fiction. I created them during a feverish week of tweaking prompts, making slight edits to images and then feeding them back to DALL•E. The result is this book-- a series of works created through a collaboration between the latent space of a vast computer program, and I suppose, the latent space of my own grey matter.
I know it's starting to get a little cliché with the AI generated image projects, but I'm pretty pleased with this one: A series of photo-realistic works of imagined "netsuke" sculptures, using DALL-E. I've created over a hundred, which I'm posting online (in addition to the Instagram account, there's a twitter account for those who prefer it), and collected 80 into a print book for people who are into that sort of thing.
Here's the intro from the book:
I’ve always been fascinated by figurines. As a kid, it was chess pieces and gaming figures, then Gothic boxwood miniatures from the 16th century. Later, I discovered the many exquisite forms of netsuke, the Japanese sculptural objects made famous during Japan's Edo period. Netsuke originally had a practical purpose: They were used to attach a box or pouch onto clothes that lacked pockets, but their aesthetic qualities, both in the diversity of subject matter and the quality of craftsmanship, were always evident. Netsuke artisans experimented for over two centuries using a variety of materials (although primarily wood and ivory) to create these sublime works of art.
The images in this book were created using a new computer program called DALL•E, developed by the company OpenAI. The program takes a string of text (known as a “prompt”), such as “A cow standing in a field during a sunset” and returns an image that is the best approximation it can find to match that text. How does it work? Essentially, the program was “trained” on millions of pairs of images with text descriptions to “learn” the connections between elements of images and certain words. This technique resulted in a digital model of the billions of inter-connections between these elements. These connections are sometimes referred to as a “latent space”-- a space that contains a near infinite number of potential images.
In my early experiments with DALL•E, I stumbled upon a prompt that could, with some regularity, produce photo-realistic images of something approximating a netsuke. These are not actual netsukes, of course. Although they may look "real", these are pure works of digital fiction. I created them during a feverish week of tweaking prompts, making slight edits to images and then feeding them back to DALL•E. The result is this book-- a series of works created through a collaboration between the latent space of a vast computer program, and I suppose, the latent space of my own grey matter.
Role: creator
« Older MetaFilter Steering Committee Self-Nominations Ope... | The Kubrick Times... Newer »
posted by Balna Watya at 7:20 PM on July 28, 2022 [1 favorite]