Guida
How This Deck Was Made — The Real Process Behind a Mother-of-Pearl Digital Deck
Let’s be clear from the start: the cards on this site are not made from real shell inlay. The images were generated by an AI model, and the card borders were composited to standardize all 78 cards to the same dimensions. This article is an honest record of that process and its limitations.
What This Deck Is
It’s a set of 78 digital cards. Each file is an 832×1248 JPEG, used for on-screen display and print previews. We have never printed and sold a physical deck.
It is not a nacre lacquerware craft. No shell was ground or cut, no lacquer was applied, and no drying was managed. There is no workshop, no production notes, and no humidity logs. In this deck, 'mother-of-pearl' refers not to a material but to the visual texture of the images.
The pearlescent glow inside the cards is painted into the images themselves, and the shimmer you see when tilting the card on screen is a CSS effect. It’s not the material reflecting light; it’s the screen making it appear that way.
How the Images Were Made
For each card, I wrote a scene description in sentences and used an image generation model to create the image. I specified the figure’s pose, gaze, attire, background terrain, and light direction in the text, then selected the result that best matched the card’s meaning.
The generated image was composited into a card frame. I measured the coordinates of the inner window and cropped all 78 images to the same position and size, so even though the artwork varies per card, the borders and margins remain consistent.
When recreating the same card, I keep a record of what was changed from the previous scene description. The prompt is the production log, and the resulting image is the output of that log.
What Was Taken from Mother-of-Pearl and Dancheong
I referenced the patterns and light textures used in nacre lacquerware, especially from Tongyeong and Wonju. I didn’t replicate the craft process but translated the color flows and pattern density of a finished nacre surface under light into the images.
The five colors of dancheong (traditional Korean decorative coloring) served as a basis for dividing the card palette. I didn’t apply blue, red, yellow, white, and black directly but adjusted them to avoid oversaturation or dullness on screen.
I also drew on the animal and plant symbols from minhwa (folk painting) and the Ten Symbols of Longevity. I brought in long-used motifs like cranes, deer, turtles, pine trees, and plum blossoms as background elements for the cards.
Attire and Iconography
The figures’ clothing is based on late Joseon dynasty dress. I specified giacca corta (jacket), mantello lungo (overcoat), gat (hat), satgat (bamboo hat), braided hair, and shoes according to each card’s status and situation. Since silhouettes vary by class and occupation even within the same era, I noted the attire separately for each card.
Some cards follow the Rider-Waite-Smith iconography, while others keep the iconography but change the background and attire. The rule was simple: does that image convey the card’s meaning more accurately?
I don’t claim strict historical accuracy in attire. I reference uigwe (royal protocols), paintings, and surviving artifact photos, but when sources differ, I don’t force one as the answer; instead, I note which version was used and why. Unverified elements remain as conjectures.
Rules Actually Followed
Each card features only one symbol in the foreground. The pearlescent glow is the background, and the figure and symbol are the protagonists. For every card, the scene description specifies attire, background, and light source, and if the generated result deviates from that description, I regenerate it.
Before publication, code checks the text. It verifies word count, paragraph count, forbidden words, internal links, structured data, and the number and length of sentences in the translation. This check is a safeguard against the text drifting from the card’s meaning, not a guarantee of factual accuracy.
The card commentary and this page’s text are drafted by AI, then fact-checked and edited by the operator before publication. Card names, meanings, and translations are managed in locale-specific data files.
Limitations and Corrections
The limitations of the image generation model remain. Regenerating with the same description yields different results, and details like fingers, fabric patterns, or background elements may be distorted. That’s why each card was generated multiple times and carefully selected.
Since this is not a physical material, the colors don’t actually change with light angle. The on-screen shimmer is a visual effect, and it disappears when printed.
The initial version of this page contained a description claiming that humidity and drying times were recorded in a workshop. That was untrue, and I replaced the entire page on November 14, 2026. If you find any incorrect descriptions, please email info@aikorea24.kr, and I will verify and correct them.
Pesca tu stesso
L'immagine che hai letto nell'articolo, questa volta ponila alla tua domanda. La stessa carta, con una domanda diversa, risponde con parole diverse.