creative-prompt-matrix

Creative Prompt Matrix

A prompt-orchestration skill that turns one product/subject description into a full batch of varied, ready-to-run image generation prompts — structured by perspective dimensions so the outputs feel like a campaign, not random variations.

Input

Collect before generating (ask in one compact question if missing):

1. Subject — What is it? (product, person, concept, service)
2. Brand direction — Tone, feeling, target audience. Optional but improves output quality.
3. Platform — Midjourney / Flux (Replicate) / fal.ai / all three
4. Count — How many prompts? Default: 12 (covers the full matrix once)

The Perspective Matrix

Each prompt is a combination of one value from each dimension. This is what creates real variation — not just rephrasing the same prompt.

Dimension	Options
Environment	studio white, urban street, nature/forest, home interior, industrial, editorial void, golden hour outdoor
Composition	hero shot (centered), lifestyle context, detail/macro, action/motion, flat lay, environmental portrait
Lighting	soft diffused, dramatic side-light, backlit/rim, natural golden, harsh direct, low-key dark
Emotion/Energy	aspirational, intimate/personal, powerful/bold, playful, serene/minimal, raw/authentic
Camera feel	editorial fashion, documentary/candid, commercial clean, fine art, sports/dynamic

Generate prompts by selecting combinations that make sense together — not all 5×7 permutations, but curated pairings that produce a coherent campaign set.

Output Format

Always output in three sections:

1. Anchor / Master Prompt

One establishing prompt that defines the subject clearly. All other prompts should feel like they belong to the same campaign as this one.

[MASTER] <full prompt text>

2. Prompt Matrix

Numbered list. Each prompt includes:

• The prompt text
• Dimension tags in brackets: [env: studio] [comp: hero] [light: soft] [emotion: aspirational]
• Platform-specific parameters appended (see references/platform-params.md)

Format:

01. <prompt text> [env: X] [comp: X] [light: X] [emotion: X]
    → MJ: <prompt> --ar 4:5 --v 7 --cref <note if master ref applies>
    → Flux: <prompt> | negative: <negative prompt>
    → fal: <prompt>

3. Consistency Notes

Brief guidance on how to maintain visual consistency across the batch:

• Midjourney: --cref strategy (use master image URL once generated)
• Flux/fal: seed locking + subject description anchoring

Prompt Writing Rules

Always include in the prompt text:

• Subject description (specific, not generic)
• Key visual adjectives (material, texture, color if relevant)
• Scene/context
• Mood/lighting descriptor
• Camera/lens feel if relevant

Avoid:

• Brand names of people or trademarked products unless user explicitly provides
• Overcrowding prompts with conflicting styles
• Generic phrases like "high quality", "beautiful", "stunning" — be specific instead

For products: Lead with the product in context. Don't center the product like a catalog shot unless composition = hero.

For people: Describe physical appearance specifically (hair, build, clothing) — don't rely on names.

Consistency Strategy (read this when user asks about consistent images)

Read references/consistency-guide.md for full strategy. Short version:

• Midjourney: Generate master → use --cref [URL] in all follow-up prompts. Use --cw 0-100 to control how closely it follows (0 = face only, 100 = full character).
• Flux on Replicate: Use flux-dev with fixed seed. Describe character/product in identical language across all prompts.
• fal.ai: Use fal-ai/flux/dev with seed parameter locked. Works well with IP-Adapter for face/product consistency.

Quick Examples

Input: "Träningsskor, sportig kvinna, 25-35 år, energisk varumärkeskänsla"

Output excerpt:

[MASTER] Athletic woman, mid-20s, strong build, wearing white performance running shoes, 
dynamic pose on urban street, morning light, editorial sports photography

01. White performance running shoes on feet of athletic woman mid-stride, urban concrete, 
    morning rim light, motion blur background [env: urban] [comp: action] [light: backlit] [emotion: powerful]
    → MJ: Athletic woman mid-stride white running shoes urban street morning rim light --ar 9:16 --v 7 --style raw
    → Flux: Athletic woman mid-stride, white performance sneakers, urban street, morning backlight, motion | negative: blurry face, deformed feet, extra limbs

Platform Quick Reference

See references/platform-params.md for full parameter lists per platform.

Midjourney defaults: --ar 4:5 --v 7 --style raw
Flux (Replicate) defaults: flux-dev model, steps: 28, guidance: 3.5
fal.ai defaults: fal-ai/flux/dev, num_inference_steps: 28