> ## Documentation Index
> Fetch the complete documentation index at: https://docs.nomiq.site/llms.txt
> Use this file to discover all available pages before exploring further.
# A Prompt Engineering Guide for the Nomiq AI Engine
> Learn how to write structured, high-signal prompts that guide Nomiq's AI Engine toward precise, differentiated brand identity outputs.
The words you type into Nomiq's prompt field are not just instructions — they are the raw data the AI Engine uses to calculate hex codes, select typefaces, and construct vector symbols. Because the engine performs semantic extraction before any visual generation begins, a weak prompt produces weak signals across all four sub-models simultaneously. Investing two extra minutes in a well-structured prompt consistently yields outputs that are closer to your vision on the first generation.
## Prompt Anatomy
Every high-yield Nomiq prompt contains four components. You do not need to write them in a rigid order, but every component must be present.
| Component | Purpose | Example |
| :----------------------- | :---------------------------------------------------- | :------------------------------------------ |
| **Industry** | Anchors the Strategy Model to a domain | "a B2B SaaS startup" |
| **Target Audience** | Shapes tone, vocabulary, and visual complexity | "targeting senior backend engineers" |
| **Emotional Adjectives** | Directly drives Color and Typography model selection | "clinical, trustworthy, precise" |
| **Specific Constraints** | Eliminates unwanted defaults and narrows visual space | "no playful gradients, use monospaced type" |
Lead with your industry and audience, then layer in emotional adjectives, then add constraints last. The Router weights earlier tokens more heavily during semantic extraction, so putting industry context first gives all sub-models a stronger foundation.
## Prompt Template
Use this template as a starting point for every new project. Replace each bracketed value with your own details.
```text theme={null}
A [industry/business type] [serving/targeting] [specific audience description].
The brand should feel [emotional adjective 1], [emotional adjective 2], and [emotional adjective 3].
[Optional: Positive constraint, e.g., "Use deep indigo and slate tones."]
[Optional: Negative constraint, e.g., "Avoid rounded corners and playful gradients."]
[Optional: Logo style, e.g., "The logo should be a text-only logotype with no icon symbol."]
```
## Good vs. Bad Prompts
The difference between a generic output and a tailored brand identity almost always traces back to prompt quality. Study these examples to calibrate your own prompts.
```text Bad Prompt ❌ — Coffee Shop theme={null}
A coffee shop.
```
```text Good Prompt ✅ — Coffee Shop theme={null}
A brutalist, third-wave coffee shop in Berlin targeting Gen-Z creatives
and design students. The brand should feel industrial, stark, and
quietly rebellious. Use high-contrast black and off-white. No brown
or warm-toned palettes. The logo should be a geometric, abstract symbol
— not a coffee cup or bean.
```
The bad prompt provides only an industry keyword. The engine defaults to its most statistically common training outputs for "coffee shop" — warm browns, a coffee bean icon, and a friendly rounded font. The good prompt locks in audience (Gen-Z creatives), emotional tone (industrial, rebellious), a color direction (black and off-white), and two negative constraints that eliminate the most generic defaults.
```text Bad Prompt ❌ — Tech Startup theme={null}
A tech startup that does AI stuff.
```
```text Good Prompt ✅ — Tech Startup theme={null}
A B2B SaaS startup building an enterprise vector database for large-scale
machine learning pipelines. The target audience is senior backend
engineers and infrastructure leads at Fortune 500 companies. The aesthetic
should be clinical, highly technical, and authoritative. Use deep indigo
and cool slate tones. Avoid playful gradients, bright accent colors, and
any rounded or bubbly design elements. Use monospaced typography in the
logotype.
```
```text Bad Prompt ❌ — Fashion Brand theme={null}
A clothing brand that's cool and sustainable.
```
```text Good Prompt ✅ — Fashion Brand theme={null}
A sustainable streetwear clothing label targeting eco-conscious urban
youth aged 18–28 in major European cities. The tone should feel gritty,
earthy, and slightly anti-establishment. Use muted, desaturated natural
tones — no bright greens or "eco" clichés. The logo should be a
text-only logotype in a condensed, all-caps sans-serif. No leaf or
plant iconography.
```
```text Bad Prompt ❌ — Personal Brand theme={null}
A Twitch streamer brand.
```
```text Good Prompt ✅ — Personal Brand theme={null}
A personal brand for a competitive first-person shooter Twitch streamer
with an aggressive, high-skill playstyle. The vibe is hyper-competitive,
fast-paced, and intimidating. The palette must use vibrant neon green
on a pure black background. Use a sharp, angular display font with no
curves. The logo symbol should be geometric and weapon-inspired without
being a literal gun.
```
## Writing for Each Prompt Component
### Specifying Target Audience
Be as specific as possible about who the brand is speaking to. Generic audience labels produce generic aesthetics. The more precisely you describe your audience, the more the engine can differentiate the output.
Include demographic specifics (age range, profession, location), psychographic context (values, lifestyle), and the relationship between the brand and audience (aspirational, peer-to-peer, authoritative).
| Weak Audience Description | Strong Audience Description |
| :------------------------ | :-------------------------------------------------------------------------------- |
| "Young people" | "Urban women aged 25–35 who follow independent fashion labels" |
| "Businesses" | "CFOs at mid-market manufacturing companies with 200–1,000 employees" |
| "Gamers" | "Competitive esports players who stream on Twitch and follow pro circuits" |
| "Health-conscious people" | "Recreational marathon runners who track biometrics and prioritize sleep science" |
### Using Emotional Adjectives
Emotional adjectives are the single most direct lever you have over the Color and Typography models. The engine maps adjective clusters to visual spaces — saturations, hue families, font classifications, and weight distributions.
Use three to five adjectives. Choose adjectives that reinforce each other rather than contradict. If you use contradictory adjectives (e.g., "playful and severe"), the engine will attempt to blend them and may produce muddy or incoherent results.
| Adjective Cluster | Color Direction | Typography Direction |
| :------------------------------------- | :--------------------------------------- | :------------------------------------------------ |
| Clinical, precise, authoritative | Cool blues, grays, high contrast | Geometric sans-serif, tight tracking |
| Warm, friendly, approachable | Warm yellows, oranges, soft neutrals | Rounded sans-serif, generous leading |
| Rebellious, gritty, anti-establishment | High contrast black/white, muted accents | Condensed display, irregular weight |
| Luxurious, refined, exclusive | Monochrome with one accent, deep tones | Serif with high x-height, elegant weight contrast |
| Energetic, aggressive, bold | Saturated neons, pure black backgrounds | Sharp angular display, heavy weight |
### Adding Style Constraints
Constraints tell the engine what to eliminate. Negative constraints are often more effective than positive ones because they reduce the output space and prevent the engine from defaulting to overused visual clichés for your industry.
Add negative constraints for any visual element you can predict the engine will default to. Every industry has predictable defaults — healthcare gets blue crosses, tech gets blue gradients, eco brands get green leaves. Name and ban the clichés you want to avoid.
```text theme={null}
Constraint Examples:
Negative constraints:
- "No brown tones"
- "Avoid rounded or bubble-style typography"
- "No coffee cup or bean iconography"
- "Do not use startup blue (#0066FF range)"
- "No gradient fills in the logo"
Positive constraints:
- "Use a monochrome palette with exactly one warm accent"
- "The logo must be a text-only logotype — no symbol or icon"
- "Use a condensed, all-caps display font for the brand name"
- "Emphasize negative space in the logo composition"
```
### Avoiding Vague and Metaphorical Language
The Semantic Extraction step cannot reliably interpret abstract metaphors or pop-culture references. If you write "our brand is the Apple of sustainable packaging," the engine may literally attempt to incorporate apple imagery. Always translate your vision into concrete design language before submitting.
| Metaphorical / Vague | Concrete Design Language |
| :------------------------------------------ | :-------------------------------------------------------------------------------------- |
| "Make it feel like a warm hug" | "Use warm amber and cream tones with rounded, friendly typography" |
| "It should feel premium, like a luxury car" | "Use a monochrome palette with one metallic accent and a high-contrast serif logotype" |
| "Give it that startup energy" | "Use a bold, geometric sans-serif and a saturated single-color palette on white" |
| "Cool and chill vibes" | "Use muted, desaturated cool tones and a light-weight sans-serif with generous spacing" |
## Iterating on Your Prompt
Prompt engineering is rarely a one-shot process. When your first generation is close but not quite right, resist the urge to rewrite the entire prompt. Instead, add targeted refinements using the Iteration Prompt bar in Brand Studio.
If you like part of the output, lock those elements before issuing a refinement. Locked elements are completely ignored by subsequent iterations, so your successful outputs are preserved while the engine focuses only on what still needs work.
```mermaid theme={null}
flowchart LR
A([Initial Prompt]) --> B(Generate Version 1)
B --> C{Evaluate Results}
C -->|Too generic| D[Add emotional adjectives]
C -->|Wrong color direction| E[Add negative color constraints]
C -->|Correct — lock it| F[Lock Strategy & Visuals]
D --> B
E --> B
classDef default fill:#111,stroke:#333,stroke-width:1px,color:#fff;
classDef decision fill:#6366f1,stroke:#4f46e5,stroke-width:1px,color:#fff;
class C decision;
```
Understand how the Router processes your prompt and dispatches it to each sub-model.
Learn how to lock successful outputs and surgically refine the elements that still need work.