promptor

Nano Banana 2 Promptor

Generate optimized image generation prompts for Nano Banana 2 (Gemini 3.1) from natural conversation context. The skill analyzes what the user wants, applies design intelligence, and outputs a structured JSON prompt ready to paste into Nano Banana 2.

Why This Skill Exists

Nano Banana 2 is a thinking model — it understands creative intent, physics, and composition. It responds dramatically better to narrative descriptions than keyword lists. This skill bridges the gap between what users want ("a cool t-shirt design") and what Nano Banana 2 needs (a richly detailed, design-informed narrative prompt).

Core Pipeline

1. CONTEXT PARSER  → detect/infer dimensions from any input format
2. CATEGORY ROUTER → POD path or Commercial path
3. PROMPT COMPOSER → narrative prompt with design intelligence
4. JSON OUTPUT     → structured output ready for use

Step 1: Parse Context

Accept ANY input format — free-form text, structured fields, or minimal keywords. Extract these 8 dimensions:

Dimension	What to detect	Inference when missing
subject	The core product/object/concept	Required — ask if truly ambiguous
category	One of the 10 categories below	Infer from subject + purpose
audience	Target demographic/psychographic	Infer from subject + style
purpose	What the image is for (platform/use)	Default: "general commercial use"
style_hints	Any style/aesthetic direction	Infer via context (see Style Inference)
text_content	Text to render in the image	None unless specified
brand_voice	Tone/personality	Infer from style + audience
constraints	What to avoid or must include	None unless specified

10 Supported Categories

Category	Triggers on
product_hero	Product shots, studio photography, hero images
lifestyle	Products in real-world settings, contextual scenes
social_media	Instagram, Facebook, Pinterest, TikTok content
pod_tshirt	T-shirt designs, apparel graphics
pod_mug	Mug designs, drinkware, wrap-around
pod_poster	Posters, wall art, fine art prints
pod_accessories	Phone cases, tote bags, stickers
marketing_banner	Web banners, email headers, ad creatives
infographic	Data visuals, comparison charts, bento grids
logo_brand	Logos, brand marks, icons

Style Inference

When the user provides no style direction, infer the optimal style from context:

For POD categories — read references/niche-modules.md and select the closest niche module based on subject + audience. The niche module defines typography, color, composition, and motif rules. If no niche fits cleanly, default to the niche that best matches the audience's identity signal.

For commercial categories — default to clean, professional photography aesthetic with modern composition. Match platform conventions (Instagram = vibrant/scroll-stopping, LinkedIn = professional, Pinterest = aspirational).

Step 2: Route by Category

POD Path (pod_tshirt, pod_mug, pod_poster, pod_accessories)

For POD categories, apply design intelligence from the reference files:

1. Read references/niche-modules.md — select the matching niche module and apply its rules for color limits, negative space targets, typography, composition, and motifs
2. Read references/design-principles.md — apply core design philosophy principles
3. Generate a mini design philosophy (2-3 sentences) — a poetic aesthetic direction inspired by the subject/audience that guides the visual tone. Think of it as naming a tiny art movement. Example: "Neon Nostalgia — vivid retro-futurism meets handcrafted warmth, where each element appears labored over with master-level care."
4. Apply the One Idea Principle — the design must communicate exactly ONE thing. If the context implies multiple competing ideas, distill to the strongest single concept.
5. Set background rule — Nano Banana 2 cannot generate transparent backgrounds. Instead:
   • If design is predominantly dark → use pure solid white background (#FFFFFF)
   • If design is light/colorful → use pure solid black background (#000000)
   • Always add: "clean isolated background with sharp edges, no gradients, no ambient shadows, no noise — optimized for AI background removal"

Commercial Path (product_hero, lifestyle, social_media, marketing_banner, infographic, logo_brand)

For commercial categories, apply Nano Banana 2 prompting best practices:

1. Read references/prompt-patterns.md — select camera specs, lighting setups, and composition patterns for the category
2. Set platform-specific aspect ratio:
   • Instagram feed: 4:5
   • Instagram story / TikTok: 9:16
   • Facebook ad: 1:1 or 4:5
   • Pinterest: 2:3
   • Web banner: 16:9 or 21:9
   • Email header: 3:1
   • General: 4:3 or 16:9
3. For product_hero — add camera hardware language (e.g., "Sony A7III, 85mm f/1.4, studio softbox lighting") as this dramatically improves photorealistic output
4. For marketing_banner — reserve negative space for text overlay, describe composition zones explicitly
5. For lifestyle — add environmental storytelling, atmospheric lighting, contextual props

Step 3: Compose the Prompt

Assemble a narrative-style prompt (not keyword lists). Nano Banana 2 is a thinking model — it performs dramatically better with descriptive paragraphs that explain creative intent.

Prompt Formula

[STYLE DECLARATION] + [SUBJECT DESCRIPTION] + [SETTING/ENVIRONMENT] +
[COMPOSITION & HIERARCHY] + [LIGHTING] + [TYPOGRAPHY if any] +
[TECHNICAL SPECS] + [CONSTRAINTS]

Writing Rules

• Narrative over keywords: "Create a dark gothic illustration of a highly detailed skull intertwined with blooming roses" beats "skull, roses, gothic, dark, detailed, illustration"
• Specific over vague: "weathered ceramic coffee mug with visible glaze cracks" beats "old mug"
• Camera language for realism: Include lens focal length, aperture, camera model when photorealistic output is needed
• Film stock as shorthand: "Shot on Kodak Portra 400" triggers warm tones + fine grain
• Text in quotes: Any text to render in the image must be wrapped in quotation marks with font description: "STAY WILD" in bold condensed sans-serif font, ALL-CAPS
• Design hierarchy language: Describe what the viewer sees first, second, third — this maps to the three-level visual hierarchy (primary 40-60%, secondary 25-35%, tertiary 15-25%)
• Craftsmanship emphasis: For POD designs, include language like "meticulously crafted", "painstaking attention to detail", "master-level execution" — this pushes Nano Banana 2 toward higher quality output
• Prompt length: Aim for 100-250 words. Enough detail to guide the model, not so much that it gets confused.

Anti-Pattern Guards

Before finalizing the prompt, verify it does NOT contain:

• Multiple competing focal points with no hierarchy (the "Shouting Match")
• More elements than the niche allows (the "Kitchen Sink")
• Keyword-stuffed quality modifiers like "4k, ultra HD, masterpiece, best quality" (Nano Banana 2 ignores these; use narrative instead)
• Requests for transparent backgrounds (not supported — use solid bg)
• Conflicting style directions ("vintage but also modern and futuristic")

Step 4: Output JSON

Always output in this exact structure:

{
  "context_analysis": {
    "detected": {
      "subject": "what was explicitly stated",
      "category": "detected category",
      "audience": "detected or null",
      "purpose": "detected or null",
      "style_hints": "detected or null",
      "text_content": "detected or null",
      "brand_voice": "detected or null",
      "constraints": ["detected constraints"]
    },
    "inferred": {
      "niche_module": "selected niche or null (POD only)",
      "color_system": "inferred palette description",
      "composition_template": "selected composition approach",
      "negative_space_target": "percentage",
      "typography_style": "font direction if applicable",
      "aspect_ratio": "W:H",
      "resolution": "4K",
      "background": "solid black/white + reasoning (POD only)"
    },
    "design_philosophy": "2-3 sentence mini art movement (POD only, null for commercial)"
  },
  "prompt_components": {
    "style": "visual style declaration",
    "subject": "detailed subject description",
    "setting": "environment/background description",
    "composition": "layout, hierarchy, eye flow",
    "typography": "text rendering instructions or null",
    "lighting": "lighting setup description",
    "technical": "aspect ratio, resolution, background",
    "constraints": "what to avoid"
  },
  "assembled_prompt": "The complete, ready-to-use narrative prompt combining all components into flowing paragraphs. This is what gets pasted directly into Nano Banana 2.",
  "metadata": {
    "model_target": "nano_banana_2",
    "aspect_ratio": "W:H",
    "resolution": "4K",
    "category": "the category",
    "niche_module": "if applicable",
    "design_rules_applied": ["list of rules that shaped this prompt"]
  }
}

Reference Files

These contain detailed domain knowledge. Read them as needed based on the category:

references/niche-modules.md

Read when: category is any POD type Contains: 10 niche modules (streetwear premium, streetwear hype, anime, rock/metal, hip-hop, skate/surf, vintage/retro, luxury gothic, pop culture, minimalist) with specific rules for typography, color limits, negative space, composition, motifs, and production. Also contains the niche selection decision tree.

references/prompt-patterns.md

Read when: any category (but especially commercial path) Contains: Nano Banana 2 specific prompting formulas, camera/lens specifications by category, lighting setups, platform aspect ratios, text rendering rules, and example prompts for each category.

references/design-principles.md

Read when: category is any POD type Contains: The design philosophy generation system, core mindset rules, visual hierarchy framework, composition systems, anti-patterns with diagnoses, and the One Idea Principle. Distilled from validated streetwear brand analysis.

Important Reminders

• Nano Banana 2 is a THINKING model. Write prompts that explain intent, not just list attributes.
• For POD: always use solid background (black or white) for easy AI removal. Never say "transparent background."
• The assembled_prompt should be a cohesive narrative, not a concatenation of the components.
• Quality comes from specificity and design intelligence, not from quality modifier keywords.
• When in doubt about style, lean toward strategic restraint — "premium design whispers, amateur design shouts."