moai-media v1.2.0 | Gemini 3 Image Preview official model
Overview
Google's official AI image brand **"Nano Banana"** was redefined in Q1 2026 from the Imagen 4 family to the Gemini 3 Image Preview family.
This skill uses the latest model IDs based on the official documentation (ai.google.dev/gemini-api/docs/image-generation) spec.
Model catalog (2 official models)
| Alias | Official model ID | Purpose | Default resolution | Cost per image |
|---|---|---|---|---|
nano-banana-pro (default) |
gemini-3-pro-image-preview |
Card news, posters, ads, text rendering SOTA | 2K | $0.134 |
nano-banana-2 |
gemini-3.1-flash-image-preview |
Drafts, high-volume A/B, cost efficiency | 1K | $0.067 |
There are only two official models: Pro and 2. No additional variants (Ultra, Cheap, etc.) are provided.
Automatic promotion of legacy/removed aliases (guarantees no modification of user code):
imagen-4.0-*,imagen-3.0-*→ automatically switches to Pro or 2nano-banana,nano-banana-ultra,ultra(alias prior to v1.1.x) → automatically promoted to Procheap→ automatically promoted to 2
⚠️ Free tier unavailable: Both models cannot be called on the Free Tier; a Pay-as-you-go paid plan is required (stated in the official documentation).
For video, use the kling skill
This skill is for images only. For video generation, use the separate skill kling.
| Purpose | Skill |
|---|---|
| Images (card news, thumbnails, posters) | nano-banana (this skill) |
| Korean large typography images | ideogram |
| All video (short-form, ads, brand) | kling |
Invocation method
CLI (simple script)
python "${CLAUDE_PLUGIN_ROOT}/scripts/generate_image.py" \
"평온한 아침 책상, 커피잔과 노트북, 자연광" \
output/slide_01.png 3:4 nano-banana-pro
Python SDK (official, recommended)
from google import genai
client = genai.Client() # GEMINI_API_KEY 환경변수 자동 인식
response = client.models.generate_content(
model="gemini-3.1-flash-image-preview", # 또는 gemini-3-pro-image-preview
contents=["Create a picture of a minimal cafe interior with warm lighting"],
)
for part in response.parts:
if part.text is not None:
print(part.text)
elif part.inline_data is not None:
image = part.as_image()
image.save("generated_image.png")
REST (camelCase)
POST https://generativelanguage.googleapis.com/v1beta/models/gemini-3-pro-image-preview:generateContent
Content-Type: application/json
{
"contents": [{"role": "user", "parts": [{"text": "..."}]}],
"generationConfig": {
"responseModalities": ["TEXT", "IMAGE"],
"imageConfig": {
"aspectRatio": "3:4",
"imageSize": "2K"
}
}
}
Response: candidates[0].content.parts[].inlineData.data (base64-encoded image)
Caution — naming convention:
- Python SDK: snake_case (
response_modalities,image_config,aspect_ratio,image_size,inline_data) - REST/JS SDK: camelCase (
responseModalities,imageConfig,aspectRatio,imageSize,inlineData)
Supported aspect ratios (14 official)
1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9,
1:4, 4:1, 1:8, 8:1
- Card news:
3:4(Instagram portrait) - Instagram feed:
1:1 - Reels/Stories:
9:16 - YouTube:
16:9 - Cinematic:
21:9 - Banner:
8:1,4:1
Supported resolutions (image_size)
This skill's generate_image.pyonly automatically sets the default per model:
| Model | Applied resolution |
|---|---|
nano-banana-pro |
"2K" (~2048px) |
nano-banana-2 |
"1K" (~1024px) |
The values supported by the official API are "512", "1K", "2K", "4K", and you can change them by injecting values directly into the script (e.g., generationConfig.imageConfig.imageSize).
Prompt tips (Korean text rendering)
Nano Banana Pro has excellent Korean typography rendering quality:
- Text in double quotesclearly:
"완벽한 주말" 이라는 제목 - Font style:
깔끔한 고딕,진한 세리프,손글씨 느낌 한글 - Position instructions:
상단 중앙에 큰 글씨로,하단 우측 작게 - Line breaks:
두 줄로 나눠서 중앙 정렬 - Background separation:
파스텔 배경, 텍스트는 검정색
API key setup
- Environment variable:
GEMINI_API_KEY(recommended)- Legacy compatibility:
NANO_BANANA_API_KEYis also recognized (zero downtime for v1.0.x users)
- Legacy compatibility:
- Issuer: ai.google.dev/gemini-api/docs/api-key
- Pay-as-you-go billing activation required (cannot be called on the free tier)
Cost examples
| Scenario | Model | Resolution | Cost |
|---|---|---|---|
| 10-card news series (recommended) | nano-banana-pro | 2K | $1.34 |
| 50 A/B draft images | nano-banana-2 | 1K | $3.35 |
| 100 thumbnails in bulk | nano-banana-2 | 1K | $6.70 |
Monthly budget example: 3 portfolio brands × 1 card news per week × one month = about $16
Migration history
- v1.1.3 (2026-04-14):
nano-banana-ultraremoved → only Pro + 2, just two. Existing aliases are automatically promoted. - v1.1.2:
gemini-2.5-flash-imageremoved, two-tier system established - v1.1.1:
google-media→nano-bananarenamed (image-only), Veo 3.1 →klingmigration - v1.1.0: Imagen 4 → Gemini 3 Image Preview migration
Related skills
kling— Dedicated to video generation (short-form, ads, lip-sync)ideogram— specialized in large Korean typography (complements this skill)elevenlabs— voice/TTS (image + narration combination)fal-gateway— non-Google models such as Flux and Recraftcard-news(moai-content) — generate card news images by calling this skill