Handmade Travel Scrapbook Collage Prompt

This is a Prompt, not a location-research engine. Fill its fact card from reliable destination sources, copy the finished prompt into an image generator, and reject any picture that swaps landmarks, invents labels, duplicates an object, or falls back on a tourist stereotype. The extra preparation is what turns an attractive collage style into a reusable editorial method.

We retained six original test images from ten generation calls across Porto, Singapore, and Marrakech. A second review rejected the first Singapore pair because its generic orchid did not support the species-level observation in the record. We generated four source-corrected candidates and retained two. The reusable bracketed version in the copy box was adapted from the filled requests rather than executed word for word, so the asset is labeled Structure Checked. The saved tests demonstrate a useful control pattern, not deterministic behavior, taxonomic identification, or automatic cultural accuracy.

Why the fact card comes before the style

A broad request for “five to seven recognizable things” makes the model decide what represents a place. That can produce visually familiar shorthand, but it can also mix a city with its country, confuse nearby architecture, turn a regional food into a municipal symbol, or add an object simply because it often appears in travel posters.

The approved card moves those choices outside the model call. Give every selected item a source and a visual description. “Dom Luis I Bridge” is a name; “a large double-deck iron arch spanning the Douro” is also a shape test. “Supertrees” is a label; “tall vertical gardens with broad branching crowns” gives the reviewer something visible to inspect.

Fill a compact destination packet

Choose one dominant landmark and enough supporting fragments to create variety. More items do not automatically improve the page. Six carefully separated motifs usually create a clearer hierarchy than ten tiny references competing for attention.

CITY: [exact place]
COUNTRY: [exact country]
HERO LANDMARK: [name + defining structure]
LANDSCAPE: [one horizon or skyline, or OMIT]
LOCAL PLANT: [verified species, or OMIT]
LOCAL ANIMAL: [verified species, or OMIT]
FOOD OR DRINK: [one item or a natural small set, or OMIT]
LOCAL TRANSPORT: [specific vehicle, or OMIT]
CULTURAL OBJECTS: [one or two sourced objects, or OMIT]
ELEMENT BUDGET: [five to seven total semantic elements including the hero;
mark every unused category OMIT]
TEXT ALLOWLIST: "[TITLE]" and optionally "[COUNTRY]"
FORBIDDEN LOCAL LOOKALIKES: [likely substitutions and stereotypes]

The total matters more than the number of available fields. Select one hero and four to six supporting elements, counting each separate cultural object individually. A natural breakfast or drink grouping counts as one semantic element. If the completed card exceeds seven, remove a weaker item before generation.

Use local institutions for place facts when possible: an official garden for a landmark garden, a transport authority for a train, a heritage collection for a craft object, and a destination authority for geography or a well-established food association. A source confirms the selection; it does not give permission to copy the source image. The output should be an original illustration built from described characteristics.

The Porto test found a missing control

The Porto base prompt requested a vintage yellow electric tram as one detached supporting fragment. Both runs included the tram, but each also placed a yellow rail vehicle across the upper deck of the Dom Luis I Bridge. That addition was locally plausible—the official Porto description notes a metro line there—but it still broke our instruction to show only the selected visual set.

Repeating “do not add anything” would not identify the actual confusion. We instead added a field for named local lookalikes. For a Porto rerun, that list should explicitly forbid an extra bridge train or metro vehicle while retaining the single detached tram. This turns an observed failure into a destination-specific repair.

Use omissions to avoid forced symbolism

The template allows OMIT because every destination does not need an animal, clothing item, vehicle, skyline, and dish simultaneously. Our Marrakech packet omitted transport and an animal. That gave more space to one Koutoubia-and-medina scene, the Atlas horizon, an olive branch, a tagine-and-tea set, zellige, and a brass lantern.

Its exclusion line also banned camels, dunes, snakes, fez hats, costume, extra minarets, extra lanterns, and other unrequested shorthand. In both corrected Marrakech runs, the supplied semantic anchors and the two allowed words were present, and none of those named lookalikes appeared.

Treat generated lettering as a review item

Limit typography to the shortest useful strings. Singapore asked for SINGAPORE once. Marrakech allowed MARRAKECH and MOROCCO, each once. All four retained corrected images passed a manual character-by-character check, but four successes do not establish future accuracy.

For a paid campaign, reusable template, translation, or printed product, generate the artwork with an empty title area and add approved typography later in a layout tool. Use this replacement block:

TEXT
Leave the title area blank warm-ivory paper. Render no letters, numbers,
symbols, signatures, captions, stamps, or watermark. The approved title will
be added as editable typography after generation.

Review the semantic set, then the craft

First compare the picture with the fact card. Only after the content passes should you judge paper fibers, brush texture, overlap, palette, and negative space. A beautiful wrong bridge is still a failed destination asset.

Do not interpret “exactly six elements” as a reliable object-count control. Our Singapore and Marrakech images included the requested semantic pieces, yet also expanded leaves or environmental foliage inside the illustrated scenes. We accepted those as non-semantic background texture while recording literal count compliance as false. If strict inventory is a legal, scientific, or catalog requirement, assemble approved objects non-generatively.

Species names require an additional evidence boundary. The retained Singapore images show the supplied orchid flower-part and color cues and the supplied long-bodied otter profile. That supports a manual morphology check against the prompt; it does not establish taxonomic identification, local expert approval, or biological accuracy.

Make one focused correction

When most of an output works, name the one failure and repeat the invariants. If the image has drifted in several ways, return to the filled fact card and generate again rather than compounding edits.

Remove only [UNAPPROVED OBJECT]. Keep the landmark silhouette, every approved
fact-card element, title spelling, paper layout, palette, torn edges, watercolor
texture, ink contours, and open negative space unchanged. Add nothing else.

The public test record preserves the exact destination prompts, file hashes, dimensions, and review decisions for all six images. For the research and approval sequence behind the prompt, continue with How to Make an AI Travel Scrapbook Collage Without Mixing Up the Destination.