generate-image

generate-image

Shared image generation skill. Two surfaces: natural language (Claude runs inline code) and script (deterministic CLI for other skills).

Models

Nickname	Model ID	When to use
nano banana 2	gemini-3.1-flash-image-preview	Default — always use this unless explicitly asked for pro
nano banana pro	gemini-3-pro-image-preview	Only when user explicitly says "pro", "higher quality", or "nano banana pro"

"Nano banana 2" and "nano banana pro" are shorthand nicknames for these models. If the user says "nano banana", default to nano banana 2 (Flash).

Never use gemini-2.5-flash-image or any gemini-2.x / gemini-1.x image model.

Invocation

Natural language (Claude writes and runs inline code)

Use when: contextual decisions needed, multi-turn generation, prompt refinement required.

from google import genai
from google.genai import types

client = genai.Client()  # reads GEMINI_API_KEY from env

response = client.models.generate_content(
    model="gemini-3.1-flash-image-preview",
    contents=["Your prompt here"],
    config=types.GenerateContentConfig(
        response_modalities=["IMAGE"],  # or ["TEXT", "IMAGE"]
        image_config=types.ImageConfig(
            aspect_ratio="16:9",  # optional
            image_size="2K"       # optional
        ),
    )
)

for part in response.parts:
    if part.thought:
        continue
    elif part.inline_data is not None:
        part.as_image().save("output.png")
    elif part.text:
        print(part.text)

Script (deterministic, called by other skills)

python ~/.claude/skills/generate-image/scripts/generate.py \
  --prompt "..." \
  [--model gemini-3.1-flash-image-preview] \
  [--aspect-ratio 16:9] \
  [--size 2K] \
  [--modalities IMAGE] \
  [--thinking minimal] \
  [--google-search] \
  [--image-search] \
  [--input-image /path/to/source.png] \
  [--output ./output.png] \
  [--output-dir ./images]

Output: JSON to stdout — { "path": "...", "text": "...", "model": "...", "usage": {...} } Errors: message to stderr + exit code 1

Params

Flag	Values	Default
--model	gemini-3.1-flash-image-preview, gemini-3-pro-image-preview	gemini-3.1-flash-image-preview
--aspect-ratio	1:1, 16:9, 9:16, 4:3, 3:4, 2:3, 3:2, 21:9, 4:5, 5:4, 1:4, 4:1, 1:8, 8:1	model default
--size	512, 1K, 2K, 4K	1K
--modalities	IMAGE, TEXT,IMAGE	IMAGE
--thinking	minimal, High	minimal
--google-search	flag	off
--image-search	flag — Flash model only	off
--input-image	file path	none (text-to-image mode)
--output	file path	./gemini-image-{timestamp}.png
--output-dir	directory	current directory

Full reference

• Aspect ratios by use case, model limits, MIME types, streaming: read $CLAUDE_SKILL_DIR/PARAMS.md
• How other skills should invoke this: read $CLAUDE_SKILL_DIR/EXAMPLES.md

Required packages

google-genai>=1.55.0
Pillow>=10.0.0