Nano Banana Skill #

Generate, edit, and iterate on visual imagery conversationally using Google’s native Nano Banana image generation foundation models (gemini-nano-banana-2.1, gemini-3.1-flash-lite-image, gemini-3.1-flash-image, gemini-3-pro-image).

Available scripts #

• scripts/banana.py: Production CLI tool for text-to-image generation, multimodal editing, and consistency anchoring with strict model capability validation.
• scripts/test_banana.py: Test suite verifying CLI argument parsing and capability validation guards.

Trigger Conditions #

Activate this skill whenever the user asks to:

• Generate new images, sprites, tilesets, or illustrations from text prompts.
• Edit, transform, style, or combine existing images.
• Use “banana” as a verb (e.g., “please banana this image”, “banana this character into anime style”, “banana this photo into chibi style”).
• Maintain character, subject, or style consistency across generated imagery using reference images (including the bundled Dani reference anchors).
• Create 4K high-resolution visual assets or seamless extreme aspect-ratio banners (1:4, 4:1, 1:8, 8:1).

Model Selection & Capability Matrix #

Deep Technical References #

Consult dedicated reference cards in references/ for full specs, exact token counts, and pixel dimensions:

• references/nano-banana-2-1.md: Consult for the primary workhorse (gemini-nano-banana-2.1), medium thinking level, seamless 1:4/4:1/1:8/8:1 panoramas, mask-based conversational editing, and Python/JS/Go SDK patterns.
• references/nano-banana-2-lite.md: Consult when building real-time UI tools, fast prototyping, or cost-critical high-frequency pipelines (1K resolution).
• references/nano-banana-2.md: Consult for gemini-3.1-flash-image 4K generation, Google Image Search Grounding, and video-to-image workflows.
• references/nano-banana-pro.md: Consult for studio asset production, multi-reference consistency across 14 anchors (6 objects + 5 characters + 3 artistic styles), and storyboards.
• references/README.md: Index and guidelines for using the bundled Dani character consistency reference anchors.

Bundled Dani Character Consistency Reference Anchors #

The references/ directory includes 4 ready-to-use PNG reference anchors for Dani that can be passed via -i / --input-image to maintain character identity or transfer visual style across scenes:

• references/chibi-dani.png: Full-body 2D chibi illustration (primary character identity & proportion anchor).
• references/celebrating-dani.png: Expressive celebratory pose (facial expression & color palette anchor).
• references/speaker-dani.png: Conference speaker pose at podium (wardrobe & upper-body angle anchor).
• references/chibi_dani_manga.png: Monochrome manga ink illustration (line-art / manga style transfer anchor).

# Multi-reference character consistency across 3 Dani anchors with Nano Banana 2.1
uv run scripts/banana.py \
  -p "banana this character: render Chibi Dani as a 16-bit pixel-art space captain on a starship bridge, keeping her exact glasses, purple hair, and proportions" \
  -i "references/chibi-dani.png" \
  -i "references/celebrating-dani.png" \
  -i "references/speaker-dani.png" \
  -f "dani_space_captain.png" \
  -m "nano-banana-2.1" \
  -r "2K" \
  -a "16:9" \
  --thinking-level "medium"

# Character + manga style transfer using chibi-dani.png and chibi_dani_manga.png
uv run scripts/banana.py \
  -p "Render the character from the first image coding furiously at a multi-monitor setup in the exact black-and-white manga ink style of the second image" \
  -i "references/chibi-dani.png" \
  -i "references/chibi_dani_manga.png" \
  -f "dani_manga_coder.png" \
  -m "nano-banana-2.1" \
  -r "2K" \
  -a "3:4"

Core Execution Workflows #

1. CLI Execution via scripts/banana.py #

Run scripts/banana.py with uv run to generate or edit images:

# Primary creative workhorse generation with Nano Banana 2.1 (Default) and medium thinking
uv run scripts/banana.py \
  -p "A seamless 2D side-scrolling pixel-art parallax background of a synthwave city skyline" \
  -f "synthwave_parallax.png" \
  -m "nano-banana-2.1" \
  -r "2K" \
  -a "4:1" \
  --thinking-level "medium" \
  --image-search

# High-velocity 1K generation with Nano Banana 2 Lite
uv run scripts/banana.py \
  -p "A minimalist flat illustration of a coffee cup on a wooden table" \
  -f "coffee_lite.png" \
  -m "nano-banana-2-lite" \
  -a "1:1"

# Studio-quality 4K generation with Nano Banana Pro
uv run scripts/banana.py \
  -p "An authentic architectural photograph of a modern library atrium with skylights" \
  -f "library_4k.png" \
  -m "nano-banana-pro" \
  -r "4K" \
  -a "16:9" \
  --search

# Conversational Image Editing ("banana this") with Multi-Reference Consistency
uv run scripts/banana.py \
  -p "banana this character: place the character into an astronaut suit on Mars" \
  -i "references/chibi-dani.png" \
  -i "references/celebrating-dani.png" \
  -f "astronaut_dani.png" \
  -m "nano-banana-2.1" \
  -r "2K"

CLI Argument Reference #

• -p, --prompt: Text prompt describing generation or edit instructions (required).
• -f, --filename: Output file path for generated PNG/JPEG (required).
• -i, --input-image: Path to input/reference image(s). Can be specified up to 14 times.
• -m, --model: nano-banana-2.1 (default, alias nano-banana-2-1), nano-banana-2, nano-banana-2-lite (alias nano-banana-lite), nano-banana-pro.
• -r, --resolution: 512px, 1K (default), 2K, 4K.
• -a, --aspect-ratio: 1:1 (default), 1:4, 1:8, 2:3, 3:2, 3:4, 4:1, 4:3, 4:5, 5:4, 8:1, 9:16, 16:9, 21:9.
• --thinking-level: minimal, medium (Nano Banana 2.1 only), or high.
• --search: Enable Google Web Search Grounding (nano-banana-2.1, nano-banana-2, nano-banana-pro).
• --image-search: Enable Google Image Search Grounding (nano-banana-2.1 and nano-banana-2).
• --api: interactions (default, Interactions API) or models (generate_content).
• --dry-run: Validate model capabilities and CLI arguments without calling the API.

2. Dual SDK Integration Patterns #

Interactions API (client.interactions.create) — Recommended #

Best for multi-turn editing, search grounding, controllable thinking, and stateful iteration:

import base64
from google import genai

client = genai.Client()

# Text-to-Image Generation with Nano Banana 2.1, 4K Resolution, Medium Thinking & Search Grounding
interaction = client.interactions.create(
    model="gemini-nano-banana-2.1",
    input="An infographic chart showing the timeline of space exploration milestones",
    tools=[{"type": "google_search", "search_types": ["web_search", "image_search"]}],
    generation_config={"thinking_level": "medium"},
    response_format={
        "type": "image",
        "aspect_ratio": "16:9",
        "image_size": "4K",
    },
)

if interaction.output_image:
    with open("space_milestones.png", "wb") as f:
        f.write(base64.b64decode(interaction.output_image.data))

Models API (client.models.generate_content) #

Direct stateless multimodal generation:

from google import genai
from PIL import Image

client = genai.Client()

img = Image.open("references/chibi-dani.png")
response = client.models.generate_content(
    model="gemini-nano-banana-2.1",
    contents=[img, "Place this character in a cozy pixel-art game developer studio with warm desk lighting."],
)

for part in response.candidates[0].content.parts:
    if part.inline_data:
        with open("dani_studio.png", "wb") as f:
            f.write(part.inline_data.data)
        break

Prompting Best Practices #

1. Be Hyper-Specific: Define materials, surface textures, lighting setups, and camera angles (three-point softbox, macro lens, shallow depth of field, 16-bit SNES pixel art).
2. Context & Intent: State the functional purpose (2D game sprite sheet with solid #FF00FF chroma key background, seamless 4:1 parallax background, app store icon).
3. Conversational Inpainting: When editing, clearly describe what to modify while instructing to preserve unchanged surroundings ("Change only the sofa to brown vintage leather. Keep all lighting and room decor untouched.").
4. Positive Framing: Describe what should appear instead of using negative constraints ("an empty street with no signs of vehicles" rather than "no cars").