One Visual Brief, Three Languages: Making AI Imagery Travel Across EN, ES and PT

If your brand publishes in more than one language, you have two consistency problems stacked on top of each other. The first is keeping imagery on-brand across a campaign. The second is keeping it on-brand across languages, where the prompt changes and the model may change its behaviour with it. The solution is to stop treating the prompt as the asset and treat the brief as the asset instead.

Why prompts do not survive translation

A prompt is a compressed instruction, and compression is exactly what translation damages. Consider one line:

"soft directional daylight from the left, gentle shadow"

Translated literally into Spanish or Portuguese, the meaning is close but the wording is longer. Longer wording competes with the rest of the prompt for influence. In practice you get images that are recognisably in the same family, with light that has drifted softer, or weaker, or warmer — enough to break a set when the three versions sit side by side.

The fix is not better translation. It is separating the layer that must not change from the layer that may.

Split the brief into a language-neutral core and a language layer

Every brief has two kinds of content:

Language-neutral core — identical in every language:

• Palette, described by material as well as hue
• Light direction, softness and temperature
• Composition and negative-space rules
• Treatment: matte, grainy, clean, dry
• The never-appear list

Language layer — rewritten natively per language:

• The subject and its job
• Mood adjectives
• Local props, setting and seasonal references
• Prompt phrasing and model-specific notes

The core is what carries brand recognition. The layer is what makes an image feel local. Keeping them in one document, clearly separated, is what allows three markets to be consistent without being identical.

Write the core once, in plain visual language

A core that translates cleanly is written as observations, not impressions. Compare:

• Hard to carry: "premium, aspirational, editorial feel"
• Easy to carry: "matte surfaces, visible grain, soft daylight from the left, 40% empty space"

The second version can be checked in a finished image and restated in any language without interpretation. This is the same principle behind writing briefs in nouns rather than adjectives, covered in the AI visual brief template. It matters more, not less, when the brief crosses a language boundary.

Verify with a three-language test

Do not assume the brief travels. Test it, with a fixed procedure:

1. Choose the least culturally specific subject you have — a plain product, a neutral surface, an object in studio light. Nothing that implies a specific country.
2. Write three prompts: one native English, one native Spanish, one native Portuguese. Each should stand on its own as natural phrasing.
3. Keep the brief's core constraints identical in all three.
4. Generate the same number of candidates for each language.
5. Place the three sets side by side on one screen and compare only four things: palette, light direction, framing and treatment.

Ignore subject detail at this stage. You are not judging whether the Spanish set is beautiful; you are judging whether it belongs to the same visual family as the English set.

Read the comparison honestly

When the sets diverge, the cause is usually one of three:

• Colour temperature drift. One set reads warmer. The light line in the core was too vague — replace "warm light" with a material and a direction: "soft daylight from the left, warm oat surfaces".
• Density drift. One set is busier. The negative-space rule was not explicit enough. Give it a number: "40% empty space, nothing within 8% of the edge".
• Treatment drift. One set looks glossier or more clinical. The treatment line lost its material words. Restore them: "matte, visible grain, dry paper backdrop".

Every one of those is a brief defect, not a language defect. Fix the core and re-run the test; the three languages usually converge on the second pass.

Localise props on purpose

Once the visual system holds, localise what the audience would actually notice: food, clothing, interiors, seasonal references, signage context. Do it as a deliberate per-language decision, written into the language layer, with the core untouched.

Two rules keep this safe:

• One local element per image, not five. A scene that changes everything reads as a different brand.
• Never localise the palette or the light. Those are the brand, and the moment they change per market you no longer have one visual identity.

Keep one canonical brief and per-language notes

A workable folder holds one canonical brief, plus a short notes file per language describing the subject phrasing, mood wording and local props used. Anyone joining the project reads the canonical brief first and the notes second.

When the brand palette changes, you update one document and three markets follow. That single source of truth is the real benefit of this structure — not the translation quality.

Publish multilingual content without duplicating work

The same principle applies to the content around the images. kublaro publishes its guides in English, Spanish and Portuguese from one canonical structure, which is why a guide such as this one exists at a single address with language variants rather than as three disconnected articles.

Start by writing your core brief, then run the three-language test with a neutral subject in the AI image generator. If the three sets look like siblings, the brief is ready for every market you serve. If they look like cousins, the core is still carrying too many adjectives.