Besties

besties [2026]

Besties is released on SuperRare on 30 June 2026, curated by jonooo.

It is a collection of 300 fully on-chain creatures, each conjured from language alone. Every artwork is generated by a large language model producing SVG code from text, with no image datasets or visual templates involved. Every path, gradient, and curve is synthesised purely through linguistic imagination. The result is a digital bestiary, a compendium of fantastical life echoing the epistemic conditions of medieval bestiary creation, where scribes and artists depicted creatures they had never seen, working only from travellers’ tales and inherited description. It also extends into a form of cyber-medievalism, where computational systems re-enact pre-modern knowledge production inside digital infrastructure.

COLLECTION
Each creature is permanently stored on the Ethereum blockchain via SSTORE2, including the SVG artwork, a scholarly bestiary entry, a traveller’s eyewitness sighting (the conceptual “prompt”), and six taxonomic traits spanning erass from 9th-century Carolingian scriptoria to the Latent Space Dark Age of 2150. No IPFS. No external servers. If every platform disappeared tomorrow, the art, text, and metadata would continue to exist fully on-chain.

The collection draws on the Sparks of AGI thesis (Bubeck et al., 2023) to position these works as capability traces: durable evidence of a text-based system converting linguistic world-models into structured visual form, committed on-chain with provenance and immutable replayability. A distinct evolutionary pressure shapes these creatures. In the medieval imagination, the monster survived because it was terrifying; in the latent space of the AI, only the cute survives. The model performs an act of radical aesthetic domestication: the basilisk’s gaze becomes a wide-eyed stare, the hydra’s teeth become a smile.

—-

In the Middle Ages, natural philosophers and artists often depicted creatures without ever seeing them firsthand. Instead, they relied on textual accounts such as travelers’ tales, oral reports, and earlier compendia of knowledge to construct visual representations of animals both real and imagined. The result was a compendium of fantastical life: dragons with serpentine tails, lions with goat heads, and birds whose plumage mirrored constellations. These works were not merely decorative; they encoded cultural, moral, and epistemic narratives, embodying the intersection of imagination, speculation, and scholarly authority.

The artworks presented here are created in a strikingly analogous fashion, yet within the framework of twenty-first-century machine intelligence. Each image emerges from a large language model tasked solely with generating SVG code in response to textual prompts. Crucially, these illustrations do not draw upon any image-based datasets or visual templates; every path, gradient, and curve is constructed purely through linguistic synthesis. In other words, these are not translations of pre-existing visuals, they are entirely original visualisations conjured by the model from language alone. This approach represents a novel mode of AI creativity: a process in which textual understanding functions as the scaffold for visual imagination, yielding artefacts that are at once singular, unique, and unrepeatable.

From a conceptual perspective, these works situate themselves within a form of digital neomedievalism. They echo the epistemic conditions of medieval bestiary creation, wherein knowledge was mediated, incomplete, and culturally encoded, but they also foreground the emerging capacities of artificial intelligence to model the world in unprecedented ways. By synthesising plausible, fantastical creatures entirely from text, these models demonstrate early steps toward more sophisticated forms of artificial general intelligence {proto-AGI}, where linguistic world-models give rise to complex, richly detailed phenomena.