Search what your pictures show.

Each file gets a position, and similar files sit together. Arcana groups them into clusters and names each one, from its own word list or from your subfolder names.

You browse by looking at regions, and a prompt lights up where its matches live.

The Alchemy tab mixes a picture with up to three materials from your collection. One of its methods works in the token grid of DINOv2, which turns a photograph into 32 × 32 tokens of 768 numbers each, and a representation autoencoder draws the mix back as a picture.

Mixed in the tokens, two materials stay one surface at every step of the way. A crossfade of their pixels shows both at once, like a double exposure.

Write a few scene descriptions. Arcana picks the image from your collection that best matches each one, in order.

Inject Poetry then passes each chosen image through a diffusion model guided by its sentence, drawing the frame closer to the words.

Inject Poetry is powered by Stability AI

Arcana offers your archive to an AI agent over MCP, as one tool that takes the same commands as Arcana's command line. The agent surveys a dataset, searches it for several ideas at once, builds a board with sections, notes and a palette, recolours pictures, and lays the board out as a page.

It edits only boards it made and copies of yours, and undo takes back each batch it sends. With the app open, its boards appear in the Moodboard tab as it works.

Given the theme disintegration, an agent built this board from the textures dataset. It follows the Bardo Thödol's dissolution of the elements, sets fragments from Artaud, Baudelaire, Dōgen, Rilke, the Heart Sutra and Ikkyū beside the photographs, and fades each spread further than the last.