Readyweights, minting concepts

We thought to experiment with a new method for sharing an AI assisted artwork and minting the concepts underlying it. In identifying our original work within the latent potential of an AI model, we introduce the idea of the readyweight (a pun on a readymade. We will explain).


On the surface this NFT involves a 3D model of an original sculpture we made, a mutant horse that appears in a larger sculpted relief we were commissioned to create for OpenAI back in 2022.

Original sketch for the relief, inspired by neural network architecture (2022)
Original sketch for the relief, inspired by neural network architecture (2022)

If you unarchive the minted 3D model (using 7zip or another tool), it contains within it some AI tools to allow for anyone to reproduce its form in infinite contexts:

  • Embeddings (concepts) compatible with multiple Stable Diffusion models

  • A LoRA and embedding for use with Stable Diffusion XL

  • Training Data so that one might train your own models on the work

  • a 3D model of the sculpture

These files are also available to play with at the following link:

Training images for readyweight
Training images for readyweight
AI images spawned from the concept of the readyweight object
AI images spawned from the concept of the readyweight object


In our 2022 Essay "Infinite Images and the latent camera" we discussed the idea of infinity related to media in this new context. Central to our argument is that any artwork now, whether unintentionally or intentionally, can serve as the seed for infinite new works.

From our perspective this new dimension of artworks, currently referred to as "training data", is something to lean into and explore deliberately. Once artists learned that distributed 2D photographs of their art were viewed more times than the 3D original, some began to modulate their works to photograph better, implicitly understanding that media recorded of the work was, for better or worse, a part of the work.

It follows that artists now might take interest in modulating their practices to consciously offer training data of original works they create, or surface and claim these concepts in latent space (more on that later). It is not our first time playing with these themes, but is perhaps the first explicitly exploring an embedding as an artwork. We feel that in the coming years it will feel increasingly familiar to see artists minting concepts in this way.

The title "Readyweight θ" refers to theta, a symbol associated with the weights of an AI model, and a cheeky nod to CC0, the license attached to all files related to this work.

So what is a Readyweight?

We have done a lot of work with embeddings over the years. For simplicities sake, embeddings are numerical representations of concepts (people, things) that exist within AI models.


For our collection Classified (Foundation, 2021) we attempted to produce portraiture of the embedding "Holly Herndon" within OpenAI's CLIP model. Self portraiture of what this model understood Holly to be.

In order to explore this concept rigorously, we enlisted the help of (future Spawning cofounders) Jordan Meyer and Patrick Hoepner. Jordan devised a technique that at the time he referred to as "Beacons" that would allow for us to train an embedding in order to hone in on what the model understood Holly to be. This was conceptually interesting, as rather than adding additional data to the model, this was more to be understood as an adventure to dig deeper into the model, an excavation of sorts. It was thrilling to see generations appear that progressively uncovered what the AI model understood of Holly.

"Holly Herndon" beacon mid-process (2021) / "Readyweight" embedding mid-process (2024)
"Holly Herndon" beacon mid-process (2021) / "Readyweight" embedding mid-process (2024)

That technique was later popularized by a paper released the following year "An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversion" by Rinon Gal et al.

What is conceptually interesting about Beacons/Textual Inversion is the idea that any original artwork made today may have already existed within the embedding space of an AI model.

It is worth sitting with that thought for a second as it is rather confounding. If I create an original sculpture, or painting, today, and train an embedding to look for that work within the latent space of a previously existing AI model, there is a strong chance it can be located as having already existed, at least conceptually, somewhere in the near infinite combination of vectors inside these models. This raises some interesting questions. We view this observation as less of a weapon to bring to existing debates over AI and copyright, and more testimony to how peculiar and fascinating this new technological substrate is for our conceptions of creativity and originality.

So, as we have established, it is possible that a new idea that we as artists have created, like an original sculpture, can be identified as having previously existed in the latent potential of an AI model. It just takes us, as artists, to excavate the model to find it, and share it, so that others might use that concept to make more works like it.

This process of an artist locating something that already exists, and minting it as an artwork, reminded us of the history of Duchamp’s readymades. As such we choose to describe these embedding artworks as readyweights.

This new sculpture of ours already existed inside embedding space, but we decided to surface and share those weights as an artwork. It is fitting to use the medium of NFTs to do so, establishing the provenance of an artwork or concept with no concern of the media associated it being used liberally by others.

Credits & Thanks

Annkathrin Kluss, Andy Rolfes, Hypopo, Jordan Meyer, Patrick Hoepner

Obviously everything to do with Readyweight θ is CC0. Do what you will.

Subscribe to Herndon Dryhurst Studio
Receive the latest updates directly to your inbox.
Mint this entry as an NFT to add it to your collection.
This entry has been permanently stored onchain and signed by its creator.