How Readings and Art Are Made

An enhanced reading is written by an AI model, not a person. You ask for one; nothing is written unless you do.

Behind it are instructions we’ve written over many tests, research into each card’s traditions, and details we calculate before a word is written, so it has less to guess.

It starts from what you wrote: your question, the cards you drew and your own words about each one. The reading builds on those words instead of replacing them. It is told to keep any look ahead conditional and not to invent details about your life.

A Woven reading is shorter and focused. A Tapestry reading is longer and draws more threads together.

A reading is something to think with. It can’t tell you what will happen.

When you draw on screen, you hold the deck and it really shuffles. It shuffles once when you press, then again about nine times a second, each time from the order before, using your device’s secure random source.

When you let go, the fan shows that exact order, one card to a place. A place keeps its card while you draw, so the card you touch is the card the shuffle put there. Whether a card is reversed is settled the same way, once for each card.

We wanted drawing on screen to work like shuffling and drawing from a real deck, not only to look like it.

Your questions, notes and readings are encrypted in our database, with a separate key for each reader. We don’t use them for advertising, and nothing you write goes into our analytics.

When you ask for an enhanced reading, the parts of it the AI needs are sent over an encrypted connection, through a routing service, to the companies hosting the models. Our setup sends it only to hosts set to zero data retention, which means they should not store it or train on it. We never use your journal to train AI models.

Some enhanced features may search the web. Our software refuses any search that contains your words. The search service sits outside the zero-data-retention setup, and no filter is perfect.

The finished reading, its sources and the searches are saved to your encrypted journal. We also keep operational records, such as which model was used and what it cost, with no reading text.

The full detail is in our Privacy Policy.

Spread Weaver has four looks for the cards, and you choose one in Settings.

• Symbolic: simple symbols drawn in code. No pictures are generated.
• Rider–Waite–Smith 1909: Pamela Colman Smith’s public-domain deck, with scans restored by Steve P.
• The Loom: 78 pictures made with an AI image model (GPT Image 2.5), following the Rider–Waite–Smith numbering, and approved by us.
• Solar Mystic: 78 pictures in a sunlit, solarpunk style, made with an AI image model (GPT Image 2.5), enlarged with Recraft Crisp, and approved by us.

The custom cards (The Loom and Solar Mystic) and the scenes on the welcome screens were all made with AI, and we have the rights to use them commercially. We directed the AI closely, tweaked the results and approved them.

Tarot readers sometimes arrive with something difficult going on. How we handle that is on our safety page.