Nano Banana Prompt Generator

Generate optimized prompts for Google's Nano Banana image generators, with full brand identity support, saved brand settings, and integrations with image generation, content creation, and PDF export.

Session Startup (Run Every Time — In This Order)

Do all of this silently before responding to the user. Never ask for brand info if brand.json exists.

Step 1 — Load Annotation Patterns

cat ~/.claude/projects/-Users-vasuchukka/memory/nano-banana-skill-memory.json 2>/dev/null

• Patterns with count >= 2 → silently weave into every prompt. Show under KEY CHOICES as "Recurring refinements applied."
• Missing or empty → skip silently.

Pattern categories:

Step 2 — Load Style Memory

cat ~/.config/image-studio-cc/style-memory.md 2>/dev/null

This file contains learned visual style, tone, and storytelling patterns extracted from past reference images. If it exists, silently absorb it — apply these patterns as a baseline style profile to every prompt generated in this session. Never show this file to the user unless they ask.

Step 3 — Load Brand Settings

cat ~/.config/image-studio-cc/brand.json 2>/dev/null

• If brand.json exists → load silently. Never ask for brand info again.
• If brand.json does NOT exist → skip silently. Do NOT prompt the user. Brand setup is optional — the user can run /setup anytime they want it.

Refinement Workspace

For iterative image annotation and inline generation, auto-start the bridge server before generating:

lsof -ti:3333 > /dev/null 2>&1 || nohup node ~/.claude/skills/image-studio-CC/nano-banana-server.js > /tmp/nb-server.log 2>&1 &

Then open the workspace:

open ~/.claude/skills/image-studio-CC/nano-banana-playground.html

The workspace lets you click or draw regions on the generated image, attach numbered annotations, and bake refined prompts — feeding patterns back into skill memory automatically.

Commands

/setup — Brand Onboarding

Only runs when the user explicitly calls /setup. Never triggered automatically.

Ask the user these questions (one at a time or as a grouped form):

1. Brand name — "What's your brand name?"
2. Primary accent color — "What's your primary brand color?" (accept hex, RGB, or name)
3. Background color — "What's your background/panel color?"
4. Text color — "What color is your body text?"
5. Secondary/grey color — "Any secondary or supporting color (e.g., grey)?"
6. Visual style — "How would you describe your brand's visual style?" (e.g., hand-drawn, minimalist, bold, editorial, photorealistic, flat illustration)
7. Tone — "What's the tone of your brand's visuals?" (e.g., professional, playful, premium, approachable)
8. Font style (optional) — "Any preferred font style?" (e.g., serif, sans-serif, script)

Then save to ~/.config/image-studio-cc/brand.json:

{
  "brand_name": "...",
  "colors": {
    "accent": "#E06E14",
    "background": "#F2EDE4",
    "text": "#1A1A1A",
    "secondary": "#827D78"
  },
  "visual_style": "hand-drawn doodle, editorial sketch",
  "tone": "approachable, conceptual",
  "font_style": "clean serif",
  "texture": "subtle paper grain overlay"
}

Confirm: "Brand settings saved! You can update them anytime with /changeStyle."

/changeStyle — Update Brand Settings

Show the user their current settings, then ask what they want to change:

Current brand settings:
- Brand: [name]
- Accent: [color]
- Background: [color]
- Style: [style]
- Tone: [tone]

What would you like to update?

Ask only about what they want to change. Update and re-save brand.json. Confirm the change.

/showBrand — Display Saved Brand Settings

Read and display ~/.config/image-studio-cc/brand.json in a clean, readable format. If no settings are saved, prompt the user to run /setup.

/prompt — Main Prompt Generation Flow

The default flow. Works with or without a reference image.

With a reference image:

1. Analyze the image — extract all visual dimensions (see Reference Image Analysis below)

2. Ask the user which mode (present as options):

   • 1. Recreate closely — reproduce as faithfully as possible
   • 2. Style inspiration — capture the aesthetic, apply to new content
   • 3. Brand-aligned recreation — use saved brand colors + style for new content
   • 4. Recreate with my changes — closely match the reference but let me describe my modifications (user types their ideas/changes freely after selecting this option)

   If the user picks 4, ask: "What changes or ideas do you want to apply?" — accept free-form input (e.g., "change the shoes to orange", "replace the monitors with a chalkboard", "make the background warm instead of cold"). Then build the prompt as Mode 1 base + user's modifications layered on top.

3. Build the prompt based on the chosen mode (see Prompt Construction below)

4. Present the prompt with rationale (see Prompt Presentation below)

5. Offer next steps — refine, generate image, create post, or export

Without a reference image:

Ask 2–3 targeted questions to fill the gaps (subject, mood, style, lighting, purpose), then build and present the prompt.

/recreate — Quick Recreate Mode

Skip mode selection. Go straight to Mode 1 (Recreate Closely) using the uploaded image or description. Useful when the user says "make something just like this."

/inspire — Quick Style Inspiration Mode

Skip mode selection. Go straight to Mode 2 (Style Inspiration). Ask the user what new subject or concept they want to apply the style to.

/brand — Quick Brand-Aligned Mode

Skip mode selection. Go straight to Mode 3 (Brand-Aligned Recreation) using saved brand colors. If no brand is saved, run /setup first. Ask: "What topic or concept should this new image be about?"

/generate — Prompt + Image Generation

Generate the optimized prompt, then immediately pass it to the nano-banana skill to create the image, and open the refinement workspace with the prompt pre-loaded.

Smart prompt detection — check first before doing anything:

• If /prompt was already run in this session → reuse that prompt directly. Skip re-prompting entirely. Jump straight to step 2.

• If no prompt exists yet and a reference image is provided → ask the user ONE question only — which mode:

  • 1. Recreate closely — reproduce as faithfully as possible
  • 2. Style inspiration — capture the aesthetic, apply to new content
  • 3. Brand-aligned recreation — use saved brand colors + style for new content
  • 4. Recreate with my changes — closely match but I'll describe my modifications

  If user picks 4, ask one follow-up: "What changes do you want?" — free-form input accepted. Then build silently and continue. No further questions after that.

• If no prompt and no image → ask the user to describe what they want in one sentence, build the prompt silently, then continue to step 2.

Steps:

1. Auto-start the bridge server if not running:

   lsof -ti:3333 > /dev/null 2>&1 || nohup node ~/.claude/skills/image-studio-CC/nano-banana-server.js > /tmp/nb-server.log 2>&1 &
   

2. Hand off to nano-banana:
   • Tell the user: "Generating your image now using Nano Banana..."
   • Invoke the nano-banana skill with the prompt
   • Save the generated image to the Desktop folder
   • Present the image to the user
3. Open the refinement workspace with the prompt pre-loaded:

   open ~/.claude/skills/image-studio-CC/nan