PromptsOct 1, 2026

Opened a Handmade Dessert Shop but Can't Build a Brand Website? Don't Worry, This Prompt Makes a Beautiful One in Three Minutes

Ask an AI to build a dessert shop website and eight times out of ten you get the same shop: pink background, rounded cards, your products redrawn as illustrations, buttons that do nothing. The tool isn't the problem. The prompt never said what must not change and what must actually work. This post gives you a condensed prompt that locks the four places that break most often, plus the three most common fixes. The full nine-section version is free in the toolbox.

#AI Website #Desserts #Bakery #Prompts #Web Design
Opened a Handmade Dessert Shop but Can't Build a Brand Website? Don't Worry, This Prompt Makes a Beautiful One in Three Minutes - Prompts

Ask an AI to build a dessert shop website and eight times out of ten you get the same shop. Pink background, rounded cards, your products redrawn as illustrations, buttons that do nothing, and a "mini-game" that is one button that wobbles when clicked. That's not because the AI can't do it. It's because you never told it what must not change and what must actually work. The condensed prompt below locks the four places that break most often; the full version I put together has nine sections, linked at the end.

TL;DR
① Say "the photos are the hero" first: hero, product cards, and product dialog all use the photos you uploaded, never illustrations or AI-generated fakes.
② The design grows out of the photos: analyze palette, silhouette, and texture first, then decide backgrounds, buttons, and motion. No default pink.
③ What should work must actually work: the cart adds, removes, merges, and totals; the game has start, input, score, end, and replay.
④ Write the language switch and the missing-glyph check into the prompt, or Chinese headings show up as boxes.
⑤ When the result is off, fix one thing at a time in the same conversation. Don't re-paste the whole brief.

Why every result looks like the same shop

Try it once and you'll see the failures all look alike. It comes down to three things:

  1. You never said the photos are the hero. Handling images is effort for an AI, so if you don't insist, it takes the cheap path: redraws your cake as an SVG illustration or grabs a generic dessert image. You wanted to show a client your products, and not one photo on the site is yours.
  2. You never said the design comes from the photos. "A cute dessert website" has a default answer in the AI's head: pink, rainbows, flowers, rounded type. Chocolate cake and fruit tart get the same treatment because it never looked at the colors in your photos.
  3. You never said "it has to actually work." Write "with a cart and a mini-game" and you get a cart icon and a button that wobbles. To the AI, that counts as "has." Spell it out: add, decrease, remove, merge identical items, subtotal, empty state; the game needs start, input, score, end, replay.

The four things this prompt locks

The condensed version only talks about four things, and each one switches off a shortcut:

  • Photos are the hero: cropping, background removal, and shadows are fine, but the product's appearance, proportions, and colors may not change, and no illustrations in its place.
  • The design grows out of the photos: analyze before designing, extract the main colors from the photos, echo the product's silhouette and texture in the decoration.
  • What should work must actually work: header, product dialog, cart, mini-game, each with a minimum bar for "working."
  • Delivery has to be verified: one HTML file, self-checked for photos, missing glyphs, language, mobile, cart, and game, preview first, explanation after.

The condensed prompt (ready to use)

Copy the whole block and send it with 3 to 10 product photos to an AI that can read images and write a website (Claude, ChatGPT, and Gemini all work; so do site builders like Lovable and Bolt). Use photos of the same product line in similar lighting.

Using my uploaded product photos, design and build a complete, fully working dessert brand shopping website, delivered as a single HTML file. Identify the products in the photos first and create a design specifically for them. Do not apply a generic dessert template. [Photos are the hero] Use my uploaded photos in the hero, product cards, and product details. Cropping, background removal, and shadows are fine, but do not replace the products with SVGs, illustrations, or AI-generated products; illustrations are decoration only. Do not invent angles, cross-sections, or fillings I did not provide. [The design grows out of the photos] Analyze the photos' colors, silhouettes, textures, and decorations first, then decide the backgrounds, cards, buttons, icons, and motion. Extract the main colors from the photos; no default pink, rainbows, or the same set of flowers. Decoration can be rich, but products, prices, and buttons must stay clear. Keep whitespace. [It has to actually work] A sticky header (with language switch, cart, and mobile menu), a photo-led hero, product cards, a product dialog that closes three ways (button, backdrop click, Escape), and a demo cart (add, decrease, remove, merge identical items, subtotal, empty state; "Review order" shows a preview only, no payment). Then build one mini-game related to the products that can genuinely be played: instructions, start, input, score, end, replay, playable with touch on mobile. [Motion] Elastic hero entrance, floating and parallax at different speeds, staggered product-card reveal, a thumbnail flying to the cart on add. Not only fade-ins, and not everything moving forever; no scroll hijacking, no autoplay sound; respect reduced-motion preferences. [Language and fonts] Traditional Chinese and English with a switch, defaulting to Traditional Chinese; switching changes every string and does not clear the cart. Chinese headings in a distinctive rounded or handwritten font, body text readable; check every character actually used, no missing-glyph boxes. Icons are original SVGs, not emoji. [No fabrication] Do not infer ingredients, allergens, or origin; do not invent reviews, sales figures, or awards. If prices are missing, use clearly labeled demonstration prices. Desktop, tablet, and mobile must all work with no horizontal overflow. When finished, actually verify the photos, glyph coverage, language switch, mobile layout, cart, and game, show me the preview first, then briefly explain the design decisions.

Paste this and the most visible difference is that the site shows your cake, the colors come from your cake, and pressing the cart button really adds one item. Get those three right and everything else is fine-tuning.

When the result is off

The first version won't be perfect, and that's normal. Fix one thing at a time in the same conversation, name the place and the symptom, and don't say "make it better." The three most common:

The products still got replaced by illustrations

Point at it directly: "The products in the hero and product cards are not the photos I uploaded. Switch back to my actual photos; cropping and background removal are fine, but appearance, proportions, and colors must match the originals. When done, tell me which photo is used in which section." That last sentence matters. It forces a photo-by-photo check.

Everything turned pink

Make it show its work before changing anything: "Re-extract the main colors from my photos, list the 3 to 5 hex codes you extracted and which part of which photo each one comes from, then rebuild the backgrounds, cards, and buttons from that palette." Once it has to list hex codes, it can't fall back on pink from memory.

The mini-game is just an animation

Give it the definition of a game: "The current mini-game only plays an animation, which doesn't count. Rebuild it as something that can actually be played: instructions, start, player input, score, end condition, replay, working with both touch and keyboard."

The full version has five more fixes: missing-glyph boxes, an incomplete language switch, horizontal overflow on mobile, too many effects, and wrong cart math.

Want the full version

The condensed prompt above is enough for a solid prototype. The full version I put together has nine sections and adds: how photos should be handled so angles don't get swapped, design directions for six kinds of pastry (cake layers, cookie embossing, croissant flakes), rules against inventing sizes and flavors for product options, the complete motion list, accessibility and performance requirements, plus preparation before you start, 8 symptom-matched fix prompts, and a pre-delivery checklist. In English and Traditional Chinese.

Get the full version free: Build Your Own Dessert Website, One Prompt, Done in Three Minutes →

Sign in and you get the link. Prompts copy in one click, English and Traditional Chinese included, yours to keep.

If you want to understand why AI-made things look AI-made at a glance, read the Style DNA reverse-engineering post next. Same logic: no adjectives, only rules and "don'ts."

FAQ

Q1: Which AI tools can I use?

Anything that accepts image uploads and outputs a runnable web page. Claude (Artifacts, Claude Design), ChatGPT, and Gemini qualify, and so do site builders like Lovable, Bolt, and v0. Free plans usually cap the number of images or the output length, and this prompt plus photos uses a fair amount of quota.

Q2: Do I really need product photos?

Yes. The whole prompt is built around "photos are the hero, the design comes from the photos." Without them the AI can only produce a generic dessert site, which is exactly the one you don't want. Phone photos are fine as long as the lighting is even and the background is clean.

Q3: Can I put the generated site live as is?

It's a prototype, not a production site. Great for a brand pitch, a new-product concept page, or confirming a style before a client project. The cart is an interactive demo with no real payment, orders, or inventory; running a shop means wiring up payments and a back office, and hosting and a domain are separate costs.

Q4: Does it work for products that aren't desserts?

Yes, with two edits. Replace the examples in the "design grows out of the photos" section with your category, and replace the mini-game examples. The rest is generic; drinks, lunch boxes, and similar products follow the same logic.

If you want to turn the prototype into a shop that actually takes orders, or need someone to wire up payments, order notifications, and a back office, get in touch.

#AI Website #Desserts #Bakery #Prompts #Web Design

Recommended reading

Turn ChatGPT into Your Personal Study Coach: 14 Must-Save Learning Prompts, from Mind Maps and the Feynman Technique to AI Quiz Games - Prompts
Prompts
Oct 24, 2026

Turn ChatGPT into Your Personal Study Coach: 14 Must-Save Learning Prompts, from Mind Maps and the Feynman Technique to AI Quiz Games

With the right prompts, ChatGPT becomes a personal study coach. This post covers 4 core learning prompts in full: /summary for key-point notes, /blueprint for a knowledge structure, /flashcard for review cards and /mindmap for mind maps, each with a copy-paste prompt and an example. 10 more advanced prompts, including the Feynman technique, quiz games and active recall, are free in the toolbox.

Don't Delete That Bad Photo: 25 ChatGPT Photo-Fix Prompts for Messy Hair, Photobombers, Dark Shots and Blur - Prompts
Prompts
Oct 18, 2026

Don't Delete That Bad Photo: 25 ChatGPT Photo-Fix Prompts for Messy Hair, Photobombers, Dark Shots and Blur

Messy hair, a stranger behind you, a face that's too dark, closed eyes, camera shake: upload the photo to ChatGPT and paste a prompt that says what to fix and what must not change. Here are the 10 most-used prompts with illustrative images and a table of which kinds of blur can be fixed; all 25 are free in the toolbox. The English names are made-up shortcuts, not official commands; the comparison images are illustrative search images, so test with your own photos.

AI Travel Photos of Yourself — No Photoshoot, and It Still Looks Like You - Prompts
Prompts
Sep 23, 2026

AI Travel Photos of Yourself — No Photoshoot, and It Still Looks Like You

Hand ChatGPT a photo of yourself and ask for a travel shot, and nine times out of ten you get someone else: bigger eyes, sharper chin, plastic skin. The tool isn't the problem — the prompt never said what must not change. This post gives you a condensed prompt that locks your face and the look first, so every new scene is just three short paragraphs. Four real results included; the full kit is free in the toolbox.

Need professionaldigital services?

From the ready-to-use Digital Toolbox to custom software development and AI integration, Zeona has you covered.

Contact me