miketromba/saas-ai-imagery

SaaS AI Imagery: Prompt Engineering for Marketing Websites

Generate cohesive, branded AI image assets for SaaS marketing websites using structured prompt engineering and style consistency techniques.

The Master Prompt Architecture

Every production-quality AI image prompt follows this formula:

SUBJECT + ENVIRONMENT + COMPOSITION + LIGHTING + STYLE + CAMERA + QUALITY + NEGATIVES

Simplified 4-Element Formula (Quick Start)

[Image Type] + [Subject] + [Background Setting] + [Style]

Example: A product illustration of a dashboard interface floating in an abstract gradient space, clean minimal glassmorphism style with soft shadows

Why Structure Matters

• Text encoders process tokens sequentially; earlier words get more attention weight
• "a red car in a forest" produces different results than "a forest with a red car"
• Lead with image type and primary subject, then layer supporting details
• Specificity beats generic quality tags ("pores visible on skin" > "highly detailed")

Prompt Block System for Brand Consistency

Define reusable blocks that get prepended/appended to every prompt. This is the single most important technique for cohesive visual identity across all generated assets.

1. Style Base Block

STYLE BASE: "[Rendering style], [camera/quality descriptor]."

Choose ONE and use everywhere:

• Photorealistic: "Professional DSLR photography, sharp detail, natural textures, realistic lighting"
• Editorial: "High-end magazine photography, styled compositions, intentional color grading"
• 3D Rendered: "Clean 3D render, soft materials, studio-lit product visualization, clay-morphic"
• Illustrative: "Digital illustration, clean lines, flat color areas, vector-like quality"
• Abstract/Surreal: "Smooth glass-like surfaces, layered translucency, subtle refraction, digital sculpture"

2. Color Direction Block

COLOR: "Color palette: [primary with hex], [secondary with hex], [neutrals].
Avoid [unwanted colors]."

Use descriptive language alongside hex codes since AI models respond to descriptions.

3. Lighting Block

LIGHTING: "[Type] from [direction], [color temp]. [Shadow quality].
No [unwanted lighting]."

| Lighting Style | Prompt Language | Brand Feel | |---|---|---| | Soft natural | "Soft diffused natural light, warm color temperature" | Approachable, warm | | Studio professional | "Professional studio lighting, three-point setup, clean shadows" | Corporate, polished | | Golden hour | "Golden hour sunlight, long warm shadows, backlit glow" | Aspirational, premium | | Flat even | "Even flat lighting, minimal shadows, bright and clean" | Modern, tech, minimal |

4. Composition Block

COMPOSITION: "[Shot type], [subject placement], [depth of field],
[perspective]. [Negative space for text overlay]."

5. Negative/Exclusion Block

EXCLUSIONS: "No: [unwanted elements, styles, artifacts]."

Always include at minimum: "No: cartoonish elements, text, watermarks, logos, cluttered backgrounds"

Assembly Order

[Subject Description] + [STYLE BASE] + [COLOR] + [LIGHTING] + [COMPOSITION] + [EXCLUSIONS]

The subject changes per image. Everything else stays constant. That is how you get consistency.

Style Consistency Techniques

Technique 1: Master Prompt Template

Create a reusable template with a {SUBJECT} variable:

"{SUBJECT}. [Your style base]. Color palette dominated by [your colors].
[Your lighting]. [Your composition]. No: [your exclusions]."

Swap only the subject for each new image. All other blocks remain identical.

Technique 2: Midjourney Style References (--sref)

• Use --sref [URL] to apply the visual style of an existing image to new generations
• Use --sref [code] with 10-digit style codes from Midjourney's style library
• Use --sw [0-1000] to control style influence strength (default 100)
• Use --sref random to discover new styles, then reuse the generated code
• Stick to ONE sref code across all images in a campaign for consistency
• Combine with --sv 7 (default) for latest style reference algorithm

Technique 3: Personalization Profiles

Midjourney personalization (--p) applies learned aesthetic preferences. Create dedicated profiles for your brand by selecting images that match your desired look.

Technique 4: Seed Locking

Use --seed [number] in Midjourney to get reproducible starting points. Same seed + similar prompt = more consistent outputs.

Technique 5: Reference Image Extraction

1. Collect 3-5 reference images that nail your desired brand look
2. Feed them to an AI and ask it to extract/articulate the style definition
3. Use that extracted description as your permanent style base block

Where to Place AI Images on a SaaS Website

See placement-guide.md for detailed placement strategies.

Quick reference:

| Section | Image Type | Purpose | |---|---|---| | Hero | Abstract 3D / surreal / metaphorical | Communicate value prop emotionally | | Features | Spot illustrations (one per feature) | Clarify each feature's benefit | | How It Works | Schematic/blueprint style | Show process flow visually | | Social Proof | Lifestyle/editorial photos | Build trust and relatability | | Pricing | Premium 3D / polished icons | Signal value, justify pricing | | CTA sections | Character-driven / aspirational | Drive action with emotion | | Blog headers | Thematic illustrations | Set context for content |

Six SaaS Illustration Styles That Convert

| Style | Best Signal | Ideal Audience | Top Placement | |---|---|---|---| | Tactile Tech (vector + organic texture) | "Human and capable" | Fintech, Security, DevOps | Hero, About | | Blueprint (monolinear, schematic) | "Precise and reliable" | Engineering, PM tools | How It Works, Features | | Spot Illustrations (small, consistent) | "Organized and clear" | All SaaS products | Feature cards, Pricing | | Abstract Surreal | "Innovative and seamless" | AI, Automation, Analytics | Hero background, Headers | | High-Contrast 3D (clay-morphic, glossy) | "Premium and polished" | Enterprise, Design tools | Hero, Pricing | | Character-Driven | "Approachable and memorable" | HR, Education, Community | Everywhere (system) |

Additional Resources

• For 30+ ready-to-use prompt examples, see prompt-examples.md
• For in-depth style consistency techniques, see style-consistency.md
• For detailed image placement strategy, see placement-guide.md

Model-Specific Tips

Midjourney: Thrives on descriptive natural language + photographic terminology. Use --ar, --stylize, --sref, --style raw for control. Words at prompt start get most weight.

DALL-E / GPT Image Gen: Favors conversational, narrative descriptions. Best for exact text rendering and spatial composition. Fewer technical camera keywords needed.

Flux: Strong at photorealism and prompt adherence. Benefits from explicit camera/lens specifications and structured prompts.

Stable Diffusion (SDXL/SD3): Requires strict syntax, strong negative prompts, and keyword weighting (keyword:1.5). Pair with ControlNet for pose/depth control.

Iteration Workflow

1. Pass 1 (Core Idea): Test subject + environment with minimal detail
2. Pass 2 (Composition): Add framing, angle, and spatial relationships
3. Pass 3 (Lighting + Style): Layer in your brand's lighting and style blocks
4. Pass 4 (Polish): Add camera specs, textures, negative prompts, parameters, run variations