Why We Publish the Exact AI Prompts We Use

On most AI art sites, a style is a button. You press it, something happens, and the prompt that actually shaped your image is never shown to you. If the result is not what you wanted, you have no vocabulary to ask for a change, so you press the button again and hope. We used to work that way too, with a polished internal recipe on one side and your typed words on the other. We removed that split.

Pick a style and the box fills with the real thing. For example, Ugly Doodle is literally this:

a funny, deliberately bad doodle drawn like a cheap marker or crayon sketch, messy and childlike with rough uneven linework, awkward proportions and warped perspective, slightly exaggerated facial features, a simple flat cartoon background of buildings, trees, or a street scrawled in quick sketchy strokes, uneven patchy coloring with visible pen strokes, looking like a lazy, silly meme-style doodle — intentionally unpolished and never realistic

And Low-Key Cinematic is this:

a moody low-key cinematic photograph on a pure black background, the subject positioned toward the right of the frame, lit with dramatic Rembrandt lighting — a single strong light illuminating half the face while the rest falls into deep shadow, stark high contrast between light and dark, a horizontal motion-blur streak trailing behind the subject to suggest fast movement, ultra-detailed 4K cinematic quality

Every word of that is yours to edit. Delete the motion blur. Move the subject left. Swap Rembrandt lighting for flat daylight. The edited text is what goes out.

We do add a thin wrapper around your text, and there is no reason to hide it either. The request is assembled as:

• The artwork's subjects: [your subject description]
• Art direction requested by the customer: "[your recipe text]"
• Keep the subject's likeness recognizable and true to the reference photo.
• Suitable for framing as a keepsake artwork.

Those last two sentences are guardrails. They are appended on the server, they are not in the editable box, and they cannot be deleted. They exist because this is a keepsake product: the result has to still look like your pet, and it has to work on a wall. Everything else is yours.

There are 12 preset recipes plus a free-text option. Each exists in English, Japanese and Chinese, and the three versions are maintained separately rather than machine translated at send time, so the Japanese recipe reads like a prompt written in Japanese, not like a translation of an English one. The current set:

• First Time with the Kitten, Ugly Doodle, Low-Key Cinematic, Naive Crayon Avatar, Editorial Crayon Doodle, Whimsical Flat Doodle
• 90s Anime Hero, Golden Rim-Light Portrait, Vintage Beach Embrace, Noir Glam Couple, Cream Knit Group Portrait, Windswept Sky Laughter
• Custom, where you describe it entirely yourself

The honest reason: the recipe was never the hard part. Anyone can copy that doodle sentence into any image model in ten seconds. What takes work is everything after it, and none of it is copyable from a text box.

• Recomputing effective DPI on the server after every move you make
• Refusing to print a layout that lands below 150 DPI
• Composing the print file to the exact canvas each product needs
• Embedding an sRGB profile so the printer does not guess at your colours
• The 48-hour window where you see the real print file before we produce anything

Keeping the prompt secret would have bought us nothing except a customer who cannot tell us what they want.