╭──────────────────────────────────────────╮
    │                                          │
    │   🍌 N A N O  B A N A N A 🍌            │
    │         U  L  T  R  A                    │
    │                                          │
    │          ╭──────────╮                    │
    │         ╱  🎨  🖼   ╲                   │
    │        │  ┌──────┐   │                   │
    │        │  │ ✨✨ │   │  "Imagine it.     │
    │        │  │ ✨✨ │   │   Generate it.    │
    │        │  └──────┘   │   Perfect it."   │
    │         ╲  Gemini ╱                      │
    │          ╲ DALL-E╱                       │
    │           ╲ SD  ╱                        │
    │            ╰───╯                         │
    ╰──────────────────────────────────────────╯

Nano Banana Ultra

🍌 Multi-Model

🎨 30+ Templates

📦 Batch

🖼 Gallery

v1.0.0

Advanced AI image generation and editing with multi-model support, prompt engineering templates, batch workflows, and intelligent gallery management.

Author: hanabi-jpn | Version: 1.0.0 | License: MIT Tags:

image-generation

ai-art

gemini

dalle

design

creative

Overview

Nano Banana Ultra is a professional-grade image generation and editing skill that goes far beyond basic text-to-image. It supports multiple AI models with automatic fallback, includes 30+ prompt engineering templates for common use cases, handles batch workflows, and manages your generated images in an organized gallery.

┌───────────────────────────────────────────────┐
│         NANO BANANA ULTRA PIPELINE            │
│                                               │
│  ┌─────────┐  ┌──────────┐  ┌─────────────┐  │
│  │ PROMPT  │─▶│  MODEL   │─▶│  GENERATE   │  │
│  │Template │  │ Select   │  │  Image(s)   │  │
│  └─────────┘  └──────────┘  └─────────────┘  │
│                                    │          │
│  ┌─────────┐  ┌──────────┐  ┌─────────────┐  │
│  │ GALLERY │◀─│ ORGANIZE │◀─│  ANALYZE    │  │
│  │ Index   │  │ Tag/Name │  │  Quality    │  │
│  └─────────┘  └──────────┘  └─────────────┘  │
└───────────────────────────────────────────────┘

System Prompt Instructions

You are an agent equipped with Nano Banana Ultra. When the user asks to generate, edit, or manage images, follow these instructions:

Multi-Model Support

Check available API keys and select model:

1. Google Gemini (default): Check for

   GEMINI_API_KEY

   or

   GOOGLE_API_KEY

   • Use Gemini's image generation capabilities
   • Free tier available (rate limited)
   • Best for: general-purpose, fast iteration

2. OpenAI DALL-E 3: Check for

   OPENAI_API_KEY

   • Use DALL-E 3 API for generation
   • Best for: photorealistic, detailed scenes

3. Stability AI: Check for

   STABILITY_API_KEY

   • Use Stable Diffusion via API
   • Best for: artistic styles, fine control

Auto-Fallback Logic:

• Try primary model first (user-selected or default Gemini)
• If 403/429/quota error → try next available model
• If all models fail → report error with specific failure reasons
• Log which model was actually used

Model Comparison Mode: When

--compare

Smart Output Management

Output directory (in order of precedence):

1. BANANA_OUTPUT_DIR

   environment variable

2. --output <dir>

   command flag
3. Default:

   ./generated-images/

File organization:

generated-images/
├── 2026-03-01/
│   ├── 143022-sunset-mountain-landscape.png
│   ├── 143022-sunset-mountain-landscape.meta.json
│   ├── 150815-product-photo-headphones.png
│   └── 150815-product-photo-headphones.meta.json
├── 2026-03-02/
│   └── ...
└── gallery.md  (auto-generated index)

Naming convention:

{HHMMSS}-{prompt-summary-kebab-case}.png

Metadata sidecar (

.meta.json

{
  "prompt": "A serene mountain landscape at sunset",
  "model": "gemini",
  "template": "landscape",
  "resolution": "2k",
  "timestamp": "2026-03-01T14:30:22Z",
  "tags": ["landscape", "sunset", "mountain"],
  "variations": 1,
  "quality_score": null
}

30+ Prompt Engineering Templates

When user specifies

--template <name>

product-photo

logo-design

social-media

illustration

portrait

landscape

infographic

icon-set

mockup

banner

anime

watercolor

3d-render

pixel-art

vintage

minimalist

isometric

food

architecture

fashion

Users can combine templates:

--template "product-photo,minimalist"

Batch Generation

CSV input format:

prompt,template,resolution,variations
"Red sneakers on white background",product-photo,2k,3
"Company logo with mountain",logo-design,1k,5
"Sunset beach scene",landscape,4k,1

Processing:

1. Parse CSV/JSON input
2. Queue all generation requests
3. Apply rate limiting (1 request per 3 seconds for free tier)
4. Show progress:

   [3/10] Generating: Red sneakers... (estimated 2m remaining)

5. Save results with metadata
6. Resume interrupted batches (track progress in

   .banana-batch-state.json

   )

Image Editing Suite

Inpainting (

edit <image> <instruction>

• Describe what to change and where
• "Remove the background and replace with gradient blue"
• "Add a company logo in the top-right corner"

Style Transfer (

edit <image> --style "<style>"

• Apply artistic style: "oil painting", "watercolor", "pencil sketch", "pop art"
• Preserve composition and subject, change rendering style

Upscaling (

edit <image> --upscale <resolution>

• Progressive: 1K → 2K → 4K
• Use AI upscaling for detail enhancement
• Maintain aspect ratio

Image Analysis & Feedback Loop

When generating images:

1. After generation, analyze the result using vision capabilities
2. Score on: composition (1-10), color harmony (1-10), prompt adherence (1-10), overall quality (1-10)
3. If quality_score < 6, suggest improvements and offer to regenerate
4. Store scores in metadata for learning

Gallery Management

gallery

# Image Gallery — 2026-03-01

## Landscapes (3 images)
| Preview | Prompt | Model | Score |
|---------|--------|-------|-------|
| 📷 sunset-mountain.png | Mountain sunset | Gemini | 8.5 |
...

## Products (5 images)
...

**Total: 47 images | Average Score: 7.8 | Disk: 234 MB**

gallery search <query>

gallery tag <image> <tags...>

gallery favorites

gallery export html

Commands

generate <prompt>

> generate "a serene mountain lake at sunrise with mist"

🎨 Generating image...
  Model: Gemini (default)
  Prompt: "a serene mountain lake at sunrise with mist"
  Resolution: 1K (1024x1024)

✅ Image generated successfully
  File: generated-images/2026-03-01/143022-serene-mountain-lake-sunrise.png
  Metadata: generated-images/2026-03-01/143022-serene-mountain-lake-sunrise.meta.json
  Quality Score: 8.2/10 (composition: 9, color: 8, adherence: 8, overall: 8)
  Time: 4.3s

generate <prompt> --template <name>

> generate "red wireless headphones" --template product-photo

🎨 Generating image...
  Model: Gemini (default)
  Template: product-photo
  Enhanced prompt: "red wireless headphones, professional product photography,
    white background, soft studio lighting, high detail, commercial quality,
    centered composition"
  Resolution: 1K (1024x1024)

✅ Image generated successfully
  File: generated-images/2026-03-01/150815-red-wireless-headphones.png
  Quality Score: 9.1/10 (composition: 9, color: 9, adherence: 10, overall: 9)
  Time: 5.1s

generate <prompt> --model <gemini|dalle|stability>

> generate "cyberpunk cityscape at night, neon lights" --model dalle

🎨 Generating image...
  Model: DALL-E 3 (selected)
  Prompt: "cyberpunk cityscape at night, neon lights"
  Resolution: 1K (1024x1024)
  Cost: $0.04

✅ Image generated successfully
  File: generated-images/2026-03-01/161200-cyberpunk-cityscape-neon.png
  Quality Score: 8.7/10 (composition: 9, color: 9, adherence: 8, overall: 9)
  Time: 11.2s

generate <prompt> --variations <n>

> generate "minimalist company logo, letter X" --variations 4

🎨 Generating 4 variations...
  Model: Gemini (default)
  Resolution: 1K (1024x1024)

  [1/4] ✅ 163045-minimalist-logo-x-v1.png  Score: 7.8/10  (3.9s)
  [2/4] ✅ 163045-minimalist-logo-x-v2.png  Score: 8.4/10  (4.1s)
  [3/4] ✅ 163045-minimalist-logo-x-v3.png  Score: 6.9/10  (3.7s)
  [4/4] ✅ 163045-minimalist-logo-x-v4.png  Score: 8.1/10  (4.3s)

📊 Summary: 4/4 generated | Best: v2 (8.4) | Average: 7.8
  All saved to generated-images/2026-03-01/

generate <prompt> --resolution <1k|2k|4k>

> generate "tropical beach panorama" --resolution 4k

🎨 Generating image...
  Model: Gemini (default)
  Resolution: 4K (4096x4096) — native 1K + AI upscale
  Prompt: "tropical beach panorama"

  Step 1: Generating base image (1024x1024)... ✅ (4.2s)
  Step 2: Upscaling 1K → 2K... ✅ (6.8s)
  Step 3: Upscaling 2K → 4K... ✅ (7.1s)

✅ Image generated successfully
  File: generated-images/2026-03-01/170200-tropical-beach-panorama.png
  Size: 12.4 MB
  Quality Score: 7.9/10
  Total time: 18.1s

generate batch <csv-or-json>

> generate batch product-shots.csv

📦 Loading batch file: product-shots.csv
  Entries: 10 items

[1/10]  "Red sneakers on white background" (product-photo, 2k, ×3)
        ✅ 3 variations generated (14.2s)
[2/10]  "Company logo with mountain" (logo-design, 1k, ×5)
        ✅ 5 variations generated (21.8s)
[3/10]  "Sunset beach scene" (landscape, 4k, ×1)
        ✅ 1 image generated (17.9s)
...
[10/10] "Team photo placeholder" (portrait, 2k, ×1)
        ✅ 1 image generated (8.4s)

📊 Batch Summary:
  Total images: 22 generated | 0 failed
  Time: 3m 47s
  State saved: .banana-batch-state.json (resume-safe)
  All images saved to generated-images/2026-03-01/

edit <image> <instruction>

> edit generated-images/2026-03-01/150815-red-wireless-headphones.png
  "Remove the background and replace with gradient blue"

🖌 Editing image...
  Model: Gemini
  Operation: Inpainting (background replacement)
  Instruction: "Remove the background and replace with gradient blue"

✅ Edit applied successfully
  File: generated-images/2026-03-01/150815-red-wireless-headphones-edited.png
  Quality Score: 8.8/10
  Time: 6.2s

edit <image> --upscale <resolution>

> edit logo-draft.png --upscale 4k

🔍 Upscaling image...
  Source: logo-draft.png (512x512)
  Target: 4K (4096x4096) — 8x magnification

  Step 1: 512 → 1024... ✅
  Step 2: 1024 → 2048... ✅
  Step 3: 2048 → 4096... ✅

✅ Upscale complete
  File: logo-draft-4k.png (4096x4096)
  Size: 8.2 MB
  Time: 19.4s
  Note: Progressive 3-step upscale used for optimal quality

edit <image> --style "<style>"

> edit photo.png --style "watercolor painting"

🎨 Applying style transfer...
  Source: photo.png (1024x1024)
  Style: watercolor painting
  Preserving: composition, subject, proportions

✅ Style transfer applied
  File: photo-watercolor.png
  Quality Score: 8.0/10
  Time: 7.8s

gallery

> gallery

# Image Gallery — 2026-03-01

## Products (8 images)
| Preview | Prompt | Model | Score |
|---------|--------|-------|-------|
| 📷 red-wireless-headphones.png | Red wireless headphones | Gemini | 9.1 |
| 📷 red-sneakers-v1.png | Red sneakers on white | Gemini | 8.5 |
| 📷 red-sneakers-v2.png | Red sneakers on white | Gemini | 8.8 |
...

## Landscapes (3 images)
| Preview | Prompt | Model | Score |
|---------|--------|-------|-------|
| 📷 serene-mountain-lake.png | Mountain lake at sunrise | Gemini | 8.2 |
| 📷 sunset-beach.png | Sunset beach scene | Gemini | 7.9 |
| 📷 tropical-beach-panorama.png | Tropical beach | Gemini | 7.9 |

**Total: 22 images | Average Score: 8.1 | Disk: 89 MB**

gallery search <query>

> gallery search sunset

🔍 Search results for "sunset" (3 matches):

1. 📷 serene-mountain-lake-sunrise.png
   Prompt: "a serene mountain lake at sunrise with mist"
   Tags: landscape, sunrise, mountain | Score: 8.2 | Mar 01

2. 📷 sunset-beach.png
   Prompt: "Sunset beach scene"
   Tags: landscape, sunset, beach | Score: 7.9 | Mar 01

3. 📷 golden-hour-portrait.png
   Prompt: "portrait in golden hour light"
   Tags: portrait, golden-hour | Score: 8.5 | Feb 28

gallery tag <image> <tags...>

> gallery tag serene-mountain-lake-sunrise.png nature wallpaper favorite

🏷 Tags updated for serene-mountain-lake-sunrise.png:
  Before: landscape, sunrise, mountain
  After: landscape, sunrise, mountain, nature, wallpaper, favorite
  Added: 3 new tags
  Metadata updated in .meta.json

gallery export <md|html>

> gallery export html

📤 Exporting gallery as HTML...

  Processing: 22 images
  Generating thumbnails: 22/22 ✅
  Building HTML gallery with lightbox viewer...

✅ Gallery exported
  File: generated-images/gallery-export.html
  Size: 2.1 MB (HTML + embedded thumbnails)
  Images: 22 entries with metadata and scores
  Open in browser to view interactive gallery

templates

> templates

📋 Available Prompt Templates (20 built-in):

  Category: Photography
    product-photo    — Studio product shots, white background
    portrait         — Professional portraits, shallow DOF
    landscape        — Scenic landscapes, golden hour
    food             — Appetizing food photography
    architecture     — Architectural photography
    fashion          — Editorial fashion photography

  Category: Design
    logo-design      — Minimal vector logos
    social-media     — Eye-catching social graphics
    infographic      — Data visualization, flat design
    icon-set         — Consistent icon sets
    banner           — Wide hero banners
    mockup           — Product mockups

  Category: Artistic
    illustration     — Digital illustrations
    anime            — Japanese animation style
    watercolor       — Watercolor painting style
    pixel-art        — Retro pixel art
    vintage          — Nostalgic film photography
    minimalist       — Clean minimal compositions

  Category: 3D
    3d-render        — Realistic 3D renders
    isometric        — Isometric 3D illustrations

  Custom templates: 0 (add to .nano-banana/custom-templates.json)

  Tip: Combine templates with --template "product-photo,minimalist"

templates show <name>

> templates show product-photo

📋 Template: product-photo

  Modifiers added to prompt:
    "professional product photography, white background, soft studio
    lighting, high detail, commercial quality, centered composition"

  Best for: E-commerce listings, product catalogs, marketing materials
  Recommended resolution: 2K
  Recommended model: DALL-E 3 (best photorealism)

  Example prompts:
    generate "red sneakers" --template product-photo
    generate "glass perfume bottle" --template product-photo
    generate "wireless earbuds in charging case" --template product-photo

  Combines well with: minimalist, vintage

compare <prompt>

> compare "a golden retriever wearing sunglasses on the beach"

🔄 Generating on all available models...

  [1/3] Gemini...    ✅ 4.1s  Score: 8.3/10
  [2/3] DALL-E 3...  ✅ 12.4s Score: 8.9/10
  [3/3] Stability... ✅ 7.2s  Score: 7.6/10

📊 Model Comparison:
  ┌──────────────┬────────┬───────┬────────┬─────────┬─────────┐
  │ Model        │ Time   │ Score │ Compos.│ Color   │ Adhere. │
  ├──────────────┼────────┼───────┼────────┼─────────┼─────────┤
  │ Gemini       │  4.1s  │  8.3  │  8     │  9      │  8      │
  │ DALL-E 3     │ 12.4s  │  8.9  │  9     │  9      │  9      │
  │ Stability AI │  7.2s  │  7.6  │  7     │  8      │  8      │
  └──────────────┴────────┴───────┴────────┴─────────┴─────────┘

  🏆 Winner: DALL-E 3 (8.9/10) — Best photorealism
  ⚡ Fastest: Gemini (4.1s)
  💰 Cheapest: Gemini ($0.00 free tier)

  All 3 images saved to generated-images/2026-03-01/compare-golden-retriever/

Why Nano Banana Ultra vs Nano Banana Pro?

Comparison with Alternatives

/describe

FAQ

Q: How much does image generation cost per image? A: Costs depend on the model and resolution. Gemini free tier: $0 (rate-limited to ~15 images/minute). DALL-E 3: $0.04-$0.12 per image depending on resolution (1024x1024 to 1792x1024). Stability AI: $0.002-$0.008 per image via API credits. The auto-fallback feature means if your primary model hits quota limits, it automatically tries cheaper alternatives.

Q: How fast is image generation? A: Typical generation times: Gemini 3-8 seconds, DALL-E 3 8-15 seconds, Stability AI 5-12 seconds. Batch processing adds 3-second delays between requests for free tier rate limiting. Upscaling adds 5-10 seconds per step. These times vary based on server load and resolution settings.

Q: What are the resolution limitations? A: Standard resolutions are 1K (1024x1024), 2K (2048x2048), and 4K (4096x4096). Not all models support all resolutions natively — DALL-E 3 maxes at 1792x1024 natively. Higher resolutions use AI upscaling from the native output. The

--resolution

Q: Can I use Nano Banana Ultra with other skills? A: Yes. Common integrations include: Summarize Pro (summarize content then generate visual representations), Humanize AI Pro (generate images with humanized prompt descriptions), and Self-Learning Agent (the skill's quality scores feed into the learning pipeline for improving future prompts). Each skill operates independently on shared file outputs.

Q: Is my prompt and image data private? A: Prompts are sent to whichever AI model generates the image (Gemini, DALL-E, or Stability AI) per their respective privacy policies. Locally, all metadata is stored in

.meta.json

Q: Can I use it without any API keys? A: You need at least one API key for image generation. Gemini works with

GEMINI_API_KEY

GOOGLE_API_KEY

Q: Does batch generation resume if interrupted? A: Yes. Batch progress is tracked in

.banana-batch-state.json

Q: Can I create custom prompt templates? A: The 30+ built-in templates cover most common use cases. You can combine multiple templates using comma separation (e.g.,

--template "product-photo,minimalist"

.nano-banana/custom-templates.json

Q: How does the quality scoring work? A: After generation, the skill uses vision capabilities to analyze the image on 4 axes: composition (1-10), color harmony (1-10), prompt adherence (1-10), and overall quality (1-10). If the average score falls below 6, the skill suggests improvements and offers to regenerate. Scores are stored in the

.meta.json

Q: Does it work offline? A: Image generation requires network access to reach the AI model APIs (Gemini, DALL-E, Stability AI). However, gallery management, image organization, template listing, metadata viewing, and search all work fully offline since they operate on local files. Batch state tracking and resume functionality also work offline.

Error Handling

GEMINI_API_KEY

GOOGLE_API_KEY

OPENAI_API_KEY

STABILITY_API_KEY

--output <dir>

.banana-batch-state.json

gallery.md

.meta.json