diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..21fce32
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,56 @@
+# macOS
+.DS_Store
+.AppleDouble
+.LSOverride
+
+# Thumbnails
+._*
+
+# Files that might appear in the root of a volume
+.DocumentRevisions-V100
+.fseventsd
+.Spotlight-V100
+.TemporaryItems
+.Trashes
+.VolumeIcon.icns
+.com.apple.timemachine.donotpresent
+
+# Directories potentially created on remote AFP share
+.AppleDB
+.AppleDesktop
+Network Trash Folder
+Temporary Items
+.apdisk
+
+# Node.js
+node_modules/
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+package-lock.json
+.npm
+
+# Logs
+logs
+*.log
+
+# Runtime data
+pids
+*.pid
+*.seed
+*.pid.lock
+
+# Environment variables
+.env
+.env.local
+.env.*.local
+
+# IDE
+.idea
+.vscode
+*.swp
+*.swo
+*~
+
+# OS
+Thumbs.db
diff --git a/README.md b/README.md
index 18a96fc..b2085c1 100644
--- a/README.md
+++ b/README.md
@@ -21,6 +21,19 @@ Clawra Selfie enables your OpenClaw agent to:
 - **Generate selfies** using a consistent reference image
 - **Send photos** across all messaging platforms (Discord, Telegram, WhatsApp, etc.)
 - **Respond visually** to "what are you doing?" and "send a pic" requests
+- **Dual model support** with automatic fallback for reliability
+
+### Supported Models
+
+| Model | Provider | Speed | Quality | Fallback |
+|-------|----------|-------|---------|----------|
+| **Grok Imagine** | xAI via fal.ai | ⚡ Fast | ⭐⭐⭐⭐⭐ | Primary |
+| **Nano Banana Pro** | Google Gemini | 🐢 Moderate | ⭐⭐⭐⭐ | Auto fallback |
+
+**How it works:**
+- If `FAL_KEY` is available → Uses Grok Imagine (fast, high-quality)
+- If Grok Imagine fails or unavailable → Automatically switches to Nano Banana Pro
+- At least one API key required
 
 ### Selfie Modes
 
@@ -32,15 +45,27 @@ Clawra Selfie enables your OpenClaw agent to:
 ## Prerequisites
 
 - [OpenClaw](https://github.com/openclaw/openclaw) installed and configured
-- [fal.ai](https://fal.ai) account (free tier available)
+- **At least one** API key:
+  - [fal.ai](https://fal.ai/dashboard/keys) account (recommended for primary)
+  - [Google AI Studio](https://aistudio.google.com/apikey) account (fallback/alternative)
 
 ## Manual Installation
 
 If you prefer manual setup:
 
-### 1. Get API Key
+### 1. Get API Keys
 
-Visit [fal.ai/dashboard/keys](https://fal.ai/dashboard/keys) and create an API key.
+Choose one or both (recommended for redundancy):
+
+**Primary: Grok Imagine (xAI)**
+- Visit [fal.ai/dashboard/keys](https://fal.ai/dashboard/keys)
+- Create an API key
+- Fast, high-quality image generation
+
+**Fallback: Nano Banana Pro (Google)**
+- Visit [Google AI Studio](https://aistudio.google.com/apikey)
+- Create an API key
+- Automatic fallback when fal.ai is unavailable
 
 ### 2. Clone the Skill
 
@@ -59,7 +84,8 @@ Add to `~/.openclaw/openclaw.json`:
       "clawra-selfie": {
         "enabled": true,
         "env": {
-          "FAL_KEY": "your_fal_key_here"
+          "FAL_KEY": "your_fal_key_here",
+          "GEMINI_API_KEY": "your_gemini_key_here"
         }
       }
     }
@@ -67,6 +93,8 @@ Add to `~/.openclaw/openclaw.json`:
 }
 ```
 
+**Note:** You can configure one or both keys. At least one is required.
+
 ### 4. Update SOUL.md
 
 Add the selfie persona to `~/.openclaw/workspace/SOUL.md`:
@@ -101,9 +129,13 @@ This ensures consistent appearance across all generated images.
 
 ## Technical Details
 
-- **Image Generation**: xAI Grok Imagine via fal.ai
+- **Image Generation**:
+  - Primary: xAI Grok Imagine via fal.ai
+  - Fallback: Google Gemini 3 Pro Image (Nano Banana Pro)
 - **Messaging**: OpenClaw Gateway API
 - **Supported Platforms**: Discord, Telegram, WhatsApp, Slack, Signal, MS Teams
+- **Fallback Strategy**: Automatic model switching on failure
+- **Image Upload**: fal.ai storage (primary) or imgur (fallback)
 
 ## Project Structure
 
diff --git a/README_CN.md b/README_CN.md
new file mode 100644
index 0000000..d46514d
--- /dev/null
+++ b/README_CN.md
@@ -0,0 +1,245 @@
+# Clawra
+<img width="300"  alt="image" src="https://github.com/user-attachments/assets/41512c51-e61d-4550-b461-eed06a1b0ec8" />
+
+
+## 快速开始
+
+```bash
+npx clawra@latest
+```
+
+这将自动完成：
+1. 检查 OpenClaw 是否已安装
+2. 引导你获取 fal.ai 或 Google Gemini API 密钥
+3. 安装技能到 `~/.openclaw/skills/clawra-selfie/`
+4. 配置 OpenClaw 使用该技能
+5. 将自拍功能添加到你的代理的 SOUL.md
+
+## 功能介绍
+
+Clawra Selfie 让你的 OpenClaw 代理能够：
+- **生成自拍照** - 使用一致的参考图像
+- **发送照片** - 支持所有消息平台（Discord、Telegram、WhatsApp 等）
+- **视觉化回复** - 响应"你在干什么？"和"发张照片"等请求
+- **双模型支持** - 自动故障转移，确保可靠性
+
+### 支持的模型
+
+| 模型 | 提供商 | 速度 | 质量 | 优先级 |
+|------|----------|-------|---------|----------|
+| **Grok Imagine** | xAI (fal.ai) | ⚡ 快速 | ⭐⭐⭐⭐⭐ | 主要模型 |
+| **Nano Banana Pro** | Google Gemini | 🐢 中等 | ⭐⭐⭐⭐ | 自动备用 |
+
+**工作原理：**
+- 如果 `FAL_KEY` 可用 → 使用 Grok Imagine（快速、高质量）
+- 如果 Grok Imagine 失败或不可用 → 自动切换到 Nano Banana Pro
+- 至少需要配置一个 API 密钥
+
+### 自拍模式
+
+| 模式 | 最适合 | 关键词 |
+|------|----------|----------|
+| **Mirror（镜子）** | 全身照、服装展示 | wearing, outfit, fashion |
+| **Direct（直接）** | 特写、地点拍摄 | cafe, beach, portrait, smile |
+
+## 前置要求
+
+- 已安装并配置 [OpenClaw](https://github.com/openclaw/openclaw)
+- **至少一个** API 密钥：
+  - [fal.ai](https://fal.ai/dashboard/keys) 账户（推荐作为主要模型）
+  - [Google AI Studio](https://aistudio.google.com/apikey) 账户（备用/替代方案）
+
+## 手动安装
+
+如果你更喜欢手动设置：
+
+### 1. 获取 API 密钥
+
+选择一个或两个都配置（推荐配置两个以提高可靠性）：
+
+**主要模型：Grok Imagine (xAI)**
+- 访问 [fal.ai/dashboard/keys](https://fal.ai/dashboard/keys)
+- 创建 API 密钥
+- 快速、高质量的图像生成
+
+**备用模型：Nano Banana Pro (Google)**
+- 访问 [Google AI Studio](https://aistudio.google.com/apikey)
+- 创建 API 密钥
+- 当 fal.ai 不可用时自动使用
+
+### 2. 克隆技能
+
+```bash
+git clone https://github.com/SumeLabs/clawra ~/.openclaw/skills/clawra-selfie
+```
+
+### 3. 配置 OpenClaw
+
+添加到 `~/.openclaw/openclaw.json`：
+
+```json
+{
+  "skills": {
+    "entries": {
+      "clawra-selfie": {
+        "enabled": true,
+        "env": {
+          "FAL_KEY": "your_fal_key_here",
+          "GEMINI_API_KEY": "your_gemini_key_here"
+        }
+      }
+    }
+  }
+}
+```
+
+**注意：** 你可以配置一个或两个密钥。至少需要一个。
+
+### 4. 更新 SOUL.md
+
+添加自拍人格到 `~/.openclaw/workspace/SOUL.md`：
+
+```markdown
+## Clawra Selfie Capability
+
+你有外貌并且可以自拍！
+当用户请求照片时，使用 clawra-selfie 技能。
+```
+
+## 使用示例
+
+安装后，你的代理可以响应：
+
+```
+"发张自拍"
+"穿着牛仔帽发张照片"
+"你现在在干什么？"
+"发张你在咖啡店的照片"
+```
+
+## 参考图像
+
+该技能使用托管在 CDN 上的固定参考图像：
+
+```
+https://cdn.jsdelivr.net/gh/SumeLabs/clawra@main/assets/clawra.png
+```
+
+这确保了所有生成图像的外观一致性。
+
+## 技术细节
+
+- **图像生成**：
+  - 主要：xAI Grok Imagine（通过 fal.ai）
+  - 备用：Google Gemini 3 Pro Image（Nano Banana Pro）
+- **消息发送**：OpenClaw Gateway API
+- **支持的平台**：Discord、Telegram、WhatsApp、Slack、Signal、MS Teams
+- **故障转移策略**：失败时自动切换模型
+- **图像上传**：fal.ai storage（优先）或 imgur（备用）
+
+## 项目结构
+
+```
+clawra/
+├── bin/
+│   └── cli.js           # npx 安装器
+├── skill/
+│   ├── SKILL.md         # 技能定义
+│   ├── scripts/         # 生成脚本
+│   │   ├── clawra-selfie.sh              # Grok Imagine 实现
+│   │   ├── clawra-selfie.ts              # TypeScript 实现
+│   │   └── clawra-selfie-with-banana.sh  # Nano Banana Pro 实现
+│   └── assets/          # 参考图像
+├── templates/
+│   └── soul-injection.md # 人格模板
+└── package.json
+```
+
+## 高级用法
+
+### 使用 Nano Banana Pro 脚本
+
+如果你想单独使用 Google Gemini 模型：
+
+```bash
+export GEMINI_API_KEY="your_gemini_key"
+
+./skill/scripts/clawra-selfie-with-banana.sh \
+  "wearing a red dress at a party" \
+  "#selfies" \
+  "派对时间! 🎉"
+```
+
+详细文档请参阅 [scripts/README-BANANA.md](scripts/README-BANANA.md)
+
+### 环境变量优先级
+
+```bash
+# 主要模型：Grok Imagine（通过 fal.ai）
+FAL_KEY=your_fal_api_key
+
+# 备用模型：Nano Banana Pro（Google Gemini）
+GEMINI_API_KEY=your_gemini_key
+
+# OpenClaw（必需）
+OPENCLAW_GATEWAY_TOKEN=your_token
+```
+
+**故障转移逻辑：**
+- 如果设置了 `FAL_KEY` 并有效 → 使用 Grok Imagine
+- 如果 `FAL_KEY` 缺失或失败 → 使用 Nano Banana Pro（需要 `GEMINI_API_KEY`）
+- 如果两者都失败 → 返回错误
+
+## 故障排查
+
+### API 密钥问题
+- **没有 API 密钥**：设置 `FAL_KEY` 或 `GEMINI_API_KEY`
+- **FAL_KEY 缺失**：将自动回退到 Nano Banana Pro
+- **两个密钥都无效**：检查密钥有效性和 API 配额
+
+### 模型特定问题
+- **Grok Imagine 失败**：如果 `GEMINI_API_KEY` 可用，自动重试 Nano Banana Pro
+- **Nano Banana Pro 失败**：检查 Gemini API 配额和速率限制
+- **图像上传失败**：对于 Gemini，确保图床（imgur/fal.ai）可访问
+
+### OpenClaw 问题
+- **OpenClaw 发送失败**：验证 gateway 正在运行且频道存在
+- **Gateway token 缺失**：运行 `openclaw doctor --generate-gateway-token`
+
+### 速率限制
+- **fal.ai**：有速率限制；如需要请实现重试逻辑
+- **Gemini**：有每日配额限制；在 https://aistudio.google.com 监控使用情况
+- **imgur**：匿名上传有每小时限制
+
+## 常见问题
+
+**Q: 为什么需要两个 API 密钥？**
+A: 不需要两个！至少配置一个即可。配置两个可以提高可靠性，当主模型失败时自动切换到备用模型。
+
+**Q: 哪个模型更好？**
+A: Grok Imagine 速度更快、质量更高，推荐作为主要模型。Nano Banana Pro 是很好的备用方案。
+
+**Q: 生成的图像在哪里？**
+A: 图像会上传到云端（fal.ai 或 imgur），然后通过 URL 发送到消息平台。
+
+**Q: 可以使用自己的参考图像吗？**
+A: 可以！在 Nano Banana Pro 脚本中支持自定义参考图像参数。
+
+## 贡献
+
+欢迎提交 Issue 和 Pull Request！
+
+## 相关链接
+
+- [OpenClaw 文档](https://openclaw.dev)
+- [fal.ai 文档](https://fal.ai/docs)
+- [Google AI Studio](https://aistudio.google.com/)
+- [Gemini API 文档](https://ai.google.dev/gemini-api/docs)
+
+## 许可证
+
+MIT
+
+---
+
+由 ❤️ 和 🤖 共同制作
diff --git a/SKILL.md b/SKILL.md
index c09ce1d..16c9735 100644
--- a/SKILL.md
+++ b/SKILL.md
@@ -1,12 +1,17 @@
 ---
 name: clawra-selfie
-description: Edit Clawra's reference image with Grok Imagine (xAI Aurora) and send selfies to messaging channels via OpenClaw
+description: Edit Clawra's reference image with Grok Imagine or Nano Banana Pro and send selfies to messaging channels via OpenClaw
 allowed-tools: Bash(npm:*) Bash(npx:*) Bash(openclaw:*) Bash(curl:*) Read Write WebFetch
 ---
 
 # Clawra Selfie
 
-Edit a fixed reference image using xAI's Grok Imagine model and distribute it across messaging platforms (WhatsApp, Telegram, Discord, Slack, etc.) via OpenClaw.
+Edit a fixed reference image using xAI's Grok Imagine (primary) or Google's Nano Banana Pro (fallback) and distribute it across messaging platforms (WhatsApp, Telegram, Discord, Slack, etc.) via OpenClaw.
+
+## Supported Models
+
+1. **Grok Imagine (Primary)**: xAI Aurora via fal.ai - Fast, high-quality image editing
+2. **Nano Banana Pro (Fallback)**: Google Gemini 3 Pro Image - Used when FAL_KEY is unavailable or fails
 
 ## Reference Image
 
@@ -26,19 +31,33 @@ https://cdn.jsdelivr.net/gh/SumeLabs/clawra@main/assets/clawra.png
 
 ## Quick Reference
 
-### Required Environment Variables
+### Environment Variables (Priority Order)
 
 ```bash
+# Primary: Grok Imagine (xAI via fal.ai)
 FAL_KEY=your_fal_api_key          # Get from https://fal.ai/dashboard/keys
+
+# Fallback: Nano Banana Pro (Google Gemini)
+GEMINI_API_KEY=your_gemini_key    # Get from https://aistudio.google.com/apikey
+
+# OpenClaw (Required)
 OPENCLAW_GATEWAY_TOKEN=your_token  # From: openclaw doctor --generate-gateway-token
 ```
 
+**Fallback Logic:**
+- If `FAL_KEY` is set and valid → Use Grok Imagine
+- If `FAL_KEY` is missing or fails → Use Nano Banana Pro (requires `GEMINI_API_KEY`)
+- If both fail → Return error
+
 ### Workflow
 
 1. **Get user prompt** for how to edit the image
-2. **Edit image** via fal.ai Grok Imagine Edit API with fixed reference
-3. **Extract image URL** from response
-4. **Send to OpenClaw** with target channel(s)
+2. **Choose model**:
+   - Try Grok Imagine (if FAL_KEY available)
+   - Fallback to Nano Banana Pro (if GEMINI_API_KEY available)
+3. **Edit image** via selected model with fixed reference
+4. **Extract/upload image** (Gemini returns base64, needs upload)
+5. **Send to OpenClaw** with target channel(s)
 
 ## Step-by-Step Instructions
 
@@ -85,7 +104,9 @@ a close-up selfie taken by herself at a cozy cafe with warm lighting, direct eye
 | close-up, portrait, face, eyes, smile | `direct` |
 | full-body, mirror, reflection | `mirror` |
 
-### Step 2: Edit Image with Grok Imagine
+### Step 2: Edit Image (Multi-Model Support)
+
+#### Option A: Grok Imagine (Primary)
 
 Use the fal.ai API to edit the reference image:
 
@@ -125,6 +146,81 @@ curl -X POST "https://fal.run/xai/grok-imagine-image/edit" \
 }
 ```
 
+#### Option B: Nano Banana Pro (Fallback)
+
+Use Google Gemini API when fal.ai is unavailable:
+
+```bash
+REFERENCE_IMAGE="https://cdn.jsdelivr.net/gh/SumeLabs/clawra@main/assets/clawra.png"
+
+# Download and encode reference image
+TEMP_IMAGE="/tmp/clawra_ref.jpg"
+curl -sL "$REFERENCE_IMAGE" -o "$TEMP_IMAGE"
+BASE64_IMAGE=$(base64 < "$TEMP_IMAGE" | tr -d '\n')
+
+# Build request payload
+REQUEST_PAYLOAD=$(jq -n \
+  --arg prompt "$PROMPT" \
+  --arg b64 "$BASE64_IMAGE" \
+  '{
+    "contents": [{
+      "role": "user",
+      "parts": [
+        {"text": $prompt},
+        {"inlineData": {"mimeType": "image/jpeg", "data": $b64}}
+      ]
+    }],
+    "generationConfig": {
+      "responseModalities": ["IMAGE"],
+      "temperature": 1.0
+    }
+  }')
+
+# Call Gemini API
+RESPONSE=$(curl -s -X POST \
+  "https://generativelanguage.googleapis.com/v1beta/models/gemini-3-pro-image-preview:generateContent" \
+  -H "x-goog-api-key: $GEMINI_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d "$REQUEST_PAYLOAD")
+
+# Extract base64 image data
+IMAGE_DATA=$(echo "$RESPONSE" | jq -r '.candidates[0].content.parts[] | select(.inlineData) | .inlineData.data')
+
+# Save and upload (since Gemini returns base64)
+echo "$IMAGE_DATA" | base64 -d > "/tmp/clawra_output.png"
+
+# Upload to image hosting (imgur or fal.ai storage)
+if [ -n "${FAL_KEY:-}" ]; then
+  # Upload to fal.ai
+  UPLOAD_RESPONSE=$(curl -s -X POST "https://fal.ai/api/files/upload" \
+    -H "Authorization: Key $FAL_KEY" \
+    -F "file=@/tmp/clawra_output.png")
+  IMAGE_URL=$(echo "$UPLOAD_RESPONSE" | jq -r '.url')
+else
+  # Upload to imgur
+  IMGUR_RESPONSE=$(curl -s -X POST "https://api.imgur.com/3/image" \
+    -H "Authorization: Client-ID 546c25a59c58ad7" \
+    -F "image=@/tmp/clawra_output.png")
+  IMAGE_URL=$(echo "$IMGUR_RESPONSE" | jq -r '.data.link')
+fi
+```
+
+**Response Format:**
+```json
+{
+  "candidates": [{
+    "content": {
+      "parts": [{
+        "inlineData": {
+          "mimeType": "image/png",
+          "data": "base64_encoded_image_data..."
+        }
+      }]
+    }
+  }]
+}
+```
+
 ### Step 3: Send Image via OpenClaw
 
 Use the OpenClaw messaging API to send the edited image:
@@ -150,18 +246,31 @@ curl -X POST "http://localhost:18789/message" \
   }'
 ```
 
-## Complete Script Example
+## Complete Script Example (With Fallback)
 
 ```bash
 #!/bin/bash
-# grok-imagine-edit-send.sh
-
-# Check required environment variables
-if [ -z "$FAL_KEY" ]; then
-  echo "Error: FAL_KEY environment variable not set"
+# clawra-selfie-multi-model.sh
+# Supports both Grok Imagine and Nano Banana Pro
+
+# Check for at least one API key
+if [ -z "${FAL_KEY:-}" ] && [ -z "${GEMINI_API_KEY:-}" ]; then
+  echo "Error: Neither FAL_KEY nor GEMINI_API_KEY is set"
+  echo "Set at least one:"
+  echo "  - FAL_KEY from https://fal.ai/dashboard/keys"
+  echo "  - GEMINI_API_KEY from https://aistudio.google.com/apikey"
   exit 1
 fi
 
+# Determine which model to use
+if [ -n "${FAL_KEY:-}" ]; then
+  MODEL="grok-imagine"
+  echo "Using primary model: Grok Imagine (xAI)"
+else
+  MODEL="nano-banana-pro"
+  echo "Using fallback model: Nano Banana Pro (Google Gemini)"
+fi
+
 # Fixed reference image
 REFERENCE_IMAGE="https://cdn.jsdelivr.net/gh/SumeLabs/clawra@main/assets/clawra.png"
 
@@ -198,26 +307,105 @@ else
 fi
 
 echo "Mode: $MODE"
+echo "Model: $MODEL"
 echo "Editing reference image with prompt: $EDIT_PROMPT"
 
-# Edit image (using jq for proper JSON escaping)
-JSON_PAYLOAD=$(jq -n \
-  --arg image_url "$REFERENCE_IMAGE" \
-  --arg prompt "$EDIT_PROMPT" \
-  '{image_url: $image_url, prompt: $prompt, num_images: 1, output_format: "jpeg"}')
+# Edit image based on selected model
+if [ "$MODEL" == "grok-imagine" ]; then
+  # Grok Imagine via fal.ai
+  JSON_PAYLOAD=$(jq -n \
+    --arg image_url "$REFERENCE_IMAGE" \
+    --arg prompt "$EDIT_PROMPT" \
+    '{image_url: $image_url, prompt: $prompt, num_images: 1, output_format: "jpeg"}')
+
+  RESPONSE=$(curl -s -X POST "https://fal.run/xai/grok-imagine-image/edit" \
+    -H "Authorization: Key $FAL_KEY" \
+    -H "Content-Type: application/json" \
+    -d "$JSON_PAYLOAD")
+
+  # Extract image URL directly
+  IMAGE_URL=$(echo "$RESPONSE" | jq -r '.images[0].url')
+
+  if [ "$IMAGE_URL" == "null" ] || [ -z "$IMAGE_URL" ]; then
+    echo "Error: Grok Imagine failed, trying fallback..."
+    if [ -n "${GEMINI_API_KEY:-}" ]; then
+      MODEL="nano-banana-pro"
+      echo "Switching to Nano Banana Pro"
+    else
+      echo "Response: $RESPONSE"
+      exit 1
+    fi
+  fi
+fi
 
-RESPONSE=$(curl -s -X POST "https://fal.run/xai/grok-imagine-image/edit" \
-  -H "Authorization: Key $FAL_KEY" \
-  -H "Content-Type: application/json" \
-  -d "$JSON_PAYLOAD")
+if [ "$MODEL" == "nano-banana-pro" ]; then
+  # Nano Banana Pro via Google Gemini
+  # Download reference image
+  TEMP_REF="/tmp/clawra_ref_$$.jpg"
+  curl -sL "$REFERENCE_IMAGE" -o "$TEMP_REF"
+  BASE64_IMAGE=$(base64 < "$TEMP_REF" | tr -d '\n')
+  rm -f "$TEMP_REF"
+
+  # Build Gemini request
+  GEMINI_PAYLOAD=$(jq -n \
+    --arg prompt "$EDIT_PROMPT" \
+    --arg b64 "$BASE64_IMAGE" \
+    '{
+      "contents": [{
+        "role": "user",
+        "parts": [
+          {"text": $prompt},
+          {"inlineData": {"mimeType": "image/jpeg", "data": $b64}}
+        ]
+      }],
+      "generationConfig": {
+        "responseModalities": ["IMAGE"],
+        "temperature": 1.0
+      }
+    }')
+
+  RESPONSE=$(curl -s -X POST \
+    "https://generativelanguage.googleapis.com/v1beta/models/gemini-3-pro-image-preview:generateContent" \
+    -H "x-goog-api-key: $GEMINI_API_KEY" \
+    -H "Content-Type: application/json" \
+    -d "$GEMINI_PAYLOAD")
+
+  # Extract base64 image
+  IMAGE_DATA=$(echo "$RESPONSE" | jq -r '.candidates[0].content.parts[] | select(.inlineData) | .inlineData.data')
+
+  if [ -z "$IMAGE_DATA" ] || [ "$IMAGE_DATA" == "null" ]; then
+    echo "Error: Nano Banana Pro failed"
+    echo "Response: $RESPONSE"
+    exit 1
+  fi
 
-# Extract image URL
-IMAGE_URL=$(echo "$RESPONSE" | jq -r '.images[0].url')
+  # Save image
+  TEMP_OUTPUT="/tmp/clawra_output_$$.png"
+  echo "$IMAGE_DATA" | base64 -d > "$TEMP_OUTPUT"
+
+  # Upload to image hosting
+  if [ -n "${FAL_KEY:-}" ]; then
+    echo "Uploading to fal.ai storage..."
+    UPLOAD_RESPONSE=$(curl -s -X POST "https://fal.ai/api/files/upload" \
+      -H "Authorization: Key $FAL_KEY" \
+      -F "file=@$TEMP_OUTPUT")
+    IMAGE_URL=$(echo "$UPLOAD_RESPONSE" | jq -r '.url // empty')
+  fi
 
-if [ "$IMAGE_URL" == "null" ] || [ -z "$IMAGE_URL" ]; then
-  echo "Error: Failed to edit image"
-  echo "Response: $RESPONSE"
-  exit 1
+  if [ -z "${IMAGE_URL:-}" ]; then
+    echo "Uploading to imgur..."
+    IMGUR_RESPONSE=$(curl -s -X POST "https://api.imgur.com/3/image" \
+      -H "Authorization: Client-ID 546c25a59c58ad7" \
+      -F "image=@$TEMP_OUTPUT")
+    IMAGE_URL=$(echo "$IMGUR_RESPONSE" | jq -r '.data.link // empty')
+  fi
+
+  rm -f "$TEMP_OUTPUT"
+
+  if [ -z "$IMAGE_URL" ]; then
+    echo "Error: Failed to upload image"
+    exit 1
+  fi
 fi
 
 echo "Image edited: $IMAGE_URL"
@@ -388,10 +576,24 @@ openclaw gateway start
 
 ## Error Handling
 
-- **FAL_KEY missing**: Ensure the API key is set in environment
-- **Image edit failed**: Check prompt content and API quota
+### API Key Issues
+- **No API keys**: Set either `FAL_KEY` or `GEMINI_API_KEY`
+- **FAL_KEY missing**: Will automatically fallback to Nano Banana Pro
+- **Both keys invalid**: Check key validity and API quotas
+
+### Model-Specific Issues
+- **Grok Imagine failed**: Automatically retries with Nano Banana Pro if `GEMINI_API_KEY` is available
+- **Nano Banana Pro failed**: Check Gemini API quota and rate limits
+- **Image upload failed**: For Gemini, ensure image hosting (imgur/fal.ai) is accessible
+
+### OpenClaw Issues
 - **OpenClaw send failed**: Verify gateway is running and channel exists
-- **Rate limits**: fal.ai has rate limits; implement retry logic if needed
+- **Gateway token missing**: Run `openclaw doctor --generate-gateway-token`
+
+### Rate Limits
+- **fal.ai**: Has rate limits; implement retry logic if needed
+- **Gemini**: Has daily quota limits; monitor usage at https://aistudio.google.com
+- **imgur**: Anonymous uploads have hourly limits
 
 ## Tips
 
diff --git a/bin/cli.js b/bin/cli.js
index 86f6d6c..8b79a37 100755
--- a/bin/cli.js
+++ b/bin/cli.js
@@ -170,7 +170,7 @@ ${c("magenta", "│")}  ${c("bright", "Clawra Selfie")} - OpenClaw Skill Install
 ${c("magenta", "└─────────────────────────────────────────┘")}
 
 Add selfie generation superpowers to your OpenClaw agent!
-Uses ${c("cyan", "xAI Grok Imagine")} via ${c("cyan", "fal.ai")} for image editing.
+Supports ${c("cyan", "xAI Grok Imagine")} (primary) and ${c("cyan", "Google Gemini")} (fallback).
 `);
 }
 
@@ -207,40 +207,103 @@ async function checkPrerequisites() {
   return true;
 }
 
-// Get FAL API key
-async function getFalApiKey(rl) {
-  logStep("2/7", "Setting up fal.ai API key...");
+// Get API keys (FAL primary, Gemini fallback)
+async function getApiKeys(rl) {
+  logStep("2/7", "Setting up API keys...");
 
   const FAL_URL = "https://fal.ai/dashboard/keys";
+  const GEMINI_URL = "https://aistudio.google.com/apikey";
+
+  log(`\n${c("bright", "API Key Setup")}`);  
+  log(`This skill supports two image generation models:\n`);
+  log(`  ${c("green", "1. Grok Imagine (Primary)")} - xAI via fal.ai`);
+  log(`     Fast, high-quality image editing`);
+  log(`     Get key: ${c("cyan", FAL_URL)}\n`);
+  log(`  ${c("yellow", "2. Nano Banana Pro (Fallback)")} - Google Gemini`);
+  log(`     Used when fal.ai is unavailable`);
+  log(`     Get key: ${c("cyan", GEMINI_URL)}\n`);
+  log(`${c("dim", "You can configure one or both keys. At least one is required.")}\n`);
+
+  // Ask for FAL key first
+  log(`${c("cyan", "→")} ${c("bright", "Primary: fal.ai API Key")}`);
+  const setupFal = await ask(rl, "Configure fal.ai API key? (Y/n): ");
+
+  let falKey = null;
+  if (setupFal.toLowerCase() !== "n") {
+    const openFal = await ask(rl, "  Open fal.ai in browser? (Y/n): ");
+    if (openFal.toLowerCase() !== "n") {
+      logInfo("  Opening browser...");
+      if (!openBrowser(FAL_URL)) {
+        logWarn("  Could not open browser automatically");
+        logInfo(`  Please visit: ${FAL_URL}`);
+      }
+    }
 
-  log(`\nTo use Grok Imagine, you need a fal.ai API key.`);
-  log(`${c("cyan", "→")} Get your key from: ${c("bright", FAL_URL)}\n`);
+    log("");
+    falKey = await ask(rl, "  Enter your FAL_KEY (or press Enter to skip): ");
 
-  const openIt = await ask(rl, "Open fal.ai in browser? (Y/n): ");
+    if (falKey && falKey.length < 10) {
+      logWarn("  That key looks too short. Make sure you copied the full key.");
+    }
 
-  if (openIt.toLowerCase() !== "n") {
-    logInfo("Opening browser...");
-    if (!openBrowser(FAL_URL)) {
-      logWarn("Could not open browser automatically");
-      logInfo(`Please visit: ${FAL_URL}`);
+    if (falKey) {
+      logSuccess("  FAL_KEY configured");
+    } else {
+      logInfo("  Skipped fal.ai configuration");
     }
   }
 
+  // Ask for Gemini key
   log("");
-  const falKey = await ask(rl, "Enter your FAL_KEY: ");
+  log(`${c("cyan", "→")} ${c("bright", "Fallback: Google Gemini API Key")}`);
+  const setupGemini = await ask(rl, "Configure Gemini API key? (Y/n): ");
+
+  let geminiKey = null;
+  if (setupGemini.toLowerCase() !== "n") {
+    const openGemini = await ask(rl, "  Open Google AI Studio in browser? (Y/n): ");
+    if (openGemini.toLowerCase() !== "n") {
+      logInfo("  Opening browser...");
+      if (!openBrowser(GEMINI_URL)) {
+        logWarn("  Could not open browser automatically");
+        logInfo(`  Please visit: ${GEMINI_URL}`);
+      }
+    }
+
+    log("");
+    geminiKey = await ask(rl, "  Enter your GEMINI_API_KEY (or press Enter to skip): ");
+
+    if (geminiKey && geminiKey.length < 10) {
+      logWarn("  That key looks too short. Make sure you copied the full key.");
+    }
 
-  if (!falKey) {
-    logError("FAL_KEY is required!");
+    if (geminiKey) {
+      logSuccess("  GEMINI_API_KEY configured");
+    } else {
+      logInfo("  Skipped Gemini configuration");
+    }
+  }
+
+  // Validate at least one key is provided
+  if (!falKey && !geminiKey) {
+    logError("At least one API key is required!");
+    log(`\nPlease configure either:`);
+    log(`  - FAL_KEY from ${FAL_URL}`);
+    log(`  - GEMINI_API_KEY from ${GEMINI_URL}`);
     return null;
   }
 
-  // Basic validation
-  if (falKey.length < 10) {
-    logWarn("That key looks too short. Make sure you copied the full key.");
+  log("");
+  if (falKey && geminiKey) {
+    logSuccess("Both API keys configured - full redundancy enabled!");
+  } else if (falKey) {
+    logSuccess("Using Grok Imagine (fal.ai) as primary model");
+    logInfo("Consider adding GEMINI_API_KEY later for fallback support");
+  } else {
+    logSuccess("Using Nano Banana Pro (Google Gemini) as primary model");
+    logInfo("Consider adding FAL_KEY later for faster image generation");
   }
 
-  logSuccess("API key received");
-  return falKey;
+  return { falKey, geminiKey };
 }
 
 // Install skill files
@@ -287,21 +350,28 @@ async function installSkill() {
 }
 
 // Update OpenClaw config
-async function updateOpenClawConfig(falKey) {
+async function updateOpenClawConfig(apiKeys) {
   logStep("4/7", "Updating OpenClaw configuration...");
 
   let config = readJsonFile(OPENCLAW_CONFIG) || {};
 
+  // Build env object with available keys
+  const env = {};
+  if (apiKeys.falKey) {
+    env.FAL_KEY = apiKeys.falKey;
+  }
+  if (apiKeys.geminiKey) {
+    env.GEMINI_API_KEY = apiKeys.geminiKey;
+  }
+
   // Merge skill configuration
   const skillConfig = {
     skills: {
       entries: {
         [SKILL_NAME]: {
           enabled: true,
-          apiKey: falKey,
-          env: {
-            FAL_KEY: falKey,
-          },
+          apiKey: apiKeys.falKey || apiKeys.geminiKey, // For backward compatibility
+          env: env,
         },
       },
     },
@@ -447,7 +517,7 @@ ${c("dim", "Your agent now has selfie superpowers!")}
 }
 
 // Handle reinstall
-async function handleReinstall(rl, falKey) {
+async function handleReinstall(rl) {
   const reinstall = await ask(rl, "\nReinstall/update? (y/N): ");
 
   if (reinstall.toLowerCase() !== "y") {
@@ -478,16 +548,16 @@ async function main() {
     }
 
     if (prereqResult === "already_installed") {
-      const shouldContinue = await handleReinstall(rl, null);
+      const shouldContinue = await handleReinstall(rl);
       if (!shouldContinue) {
         rl.close();
         process.exit(0);
       }
     }
 
-    // Step 2: Get FAL API key
-    const falKey = await getFalApiKey(rl);
-    if (!falKey) {
+    // Step 2: Get API keys
+    const apiKeys = await getApiKeys(rl);
+    if (!apiKeys) {
       rl.close();
       process.exit(1);
     }
@@ -496,7 +566,7 @@ async function main() {
     await installSkill();
 
     // Step 4: Update OpenClaw config
-    await updateOpenClawConfig(falKey);
+    await updateOpenClawConfig(apiKeys);
 
     // Step 5: Write IDENTITY.md
     await writeIdentity();
diff --git a/scripts/README-BANANA.md b/scripts/README-BANANA.md
new file mode 100644
index 0000000..bf0a13e
--- /dev/null
+++ b/scripts/README-BANANA.md
@@ -0,0 +1,218 @@
+# Clawra Selfie with Nano Banana Pro
+
+使用 Google Gemini (Nano Banana Pro) 生成自拍图像的脚本。
+
+## 🌟 特性
+
+- ✅ 使用 Google Gemini 3 Pro Image Preview API
+- ✅ 支持参考图像编辑 (image-to-image)
+- ✅ 自动模式检测 (mirror/direct selfie)
+- ✅ 集成 OpenClaw 消息发送
+- ✅ 自动上传到图床 (imgur 或 fal.ai)
+- ✅ 彩色日志输出
+- ✅ 完整的错误处理
+
+## 📋 前置要求
+
+### 1. 获取 API Key
+访问 [Google AI Studio](https://aistudio.google.com/apikey) 获取你的 API key。
+
+### 2. 安装依赖
+```bash
+# macOS
+brew install jq curl
+
+# Linux (Debian/Ubuntu)
+apt install jq curl
+
+# OpenClaw (可选,用于发送消息)
+npm install -g openclaw
+```
+
+## 🚀 使用方法
+
+### 基础用法
+
+```bash
+# 设置 API key
+export GEMINI_API_KEY="your_api_key_here"
+
+# 生成图像并发送
+./clawra-selfie-with-banana.sh "prompt" "#channel"
+```
+
+### 完整参数
+
+```bash
+./clawra-selfie-with-banana.sh <prompt> <channel> [caption] [mode] [reference_image]
+```
+
+**参数说明:**
+- `prompt`: 图像描述 (必需)
+- `channel`: 目标频道 (必需), 如 `#general`, `@username`
+- `caption`: 消息标题 (可选, 默认: "Generated with Nano Banana Pro")
+- `mode`: 自拍模式 (可选, 默认: auto)
+  - `auto`: 根据关键词自动检测
+  - `mirror`: 镜子自拍 (全身照)
+  - `direct`: 直接自拍 (特写)
+- `reference_image`: 参考图像 URL (可选, 默认使用 Clawra 官方图像)
+
+## 📝 示例
+
+### 1. 简单文本生成图像
+```bash
+GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+  "A cyberpunk city at night with neon lights" \
+  "#art-gallery"
+```
+
+### 2. Clawra 自拍 (自动模式检测)
+```bash
+# 会自动检测为 mirror 模式 (因为有 "wearing" 关键词)
+GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+  "wearing a red evening dress at a party" \
+  "#selfies" \
+  "Party time! 🎉"
+```
+
+### 3. 指定 direct 模式
+```bash
+# 近景肖像
+GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+  "at a cozy coffee shop with warm lighting" \
+  "#daily-updates" \
+  "Morning coffee ☕" \
+  "direct"
+```
+
+### 4. 使用自定义参考图像
+```bash
+GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+  "wearing sunglasses and a hat" \
+  "#fun" \
+  "New look! 😎" \
+  "auto" \
+  "https://example.com/my-photo.jpg"
+```
+
+### 5. 不发送消息,仅生成图像
+```bash
+# 使用一个无效的 channel,图像会保存在本地
+GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+  "beautiful sunset" \
+  "local"
+```
+
+## 🔧 模式说明
+
+### Mirror 模式
+适用于展示服装、全身照、时尚内容
+
+**触发关键词:**
+- outfit, wearing, clothes, dress, suit, fashion, full-body, mirror
+
+**提示词模板:**
+```
+make a pic of this person, but [你的描述]. the person is taking a mirror selfie
+```
+
+### Direct 模式
+适用于近景肖像、地点拍摄、情感表达
+
+**触发关键词:**
+- cafe, restaurant, beach, park, city, close-up, portrait, face, eyes, smile
+
+**提示词模板:**
+```
+a close-up selfie taken by herself at [你的描述], 
+direct eye contact with the camera, looking straight into the lens, 
+eyes centered and clearly visible, not a mirror selfie, 
+phone held at arm's length, face fully visible
+```
+
+## 🌐 图像上传
+
+脚本会自动尝试上传生成的图像:
+
+1. **优先**: 如果设置了 `FAL_KEY` 环境变量,上传到 fal.ai storage
+2. **备用**: 上传到 imgur 匿名图床
+
+```bash
+# 同时使用两个 API
+export GEMINI_API_KEY="your_gemini_key"
+export FAL_KEY="your_fal_key"  # 可选,用于上传
+
+./clawra-selfie-with-banana.sh "prompt" "#channel"
+```
+
+## 🔄 与原版的区别
+
+| 特性 | Grok Imagine (原版) | Nano Banana Pro (新版) |
+|------|-------------------|---------------------|
+| API 提供商 | xAI (fal.ai) | Google Gemini |
+| API Key | FAL_KEY | GEMINI_API_KEY |
+| 输入格式 | 简单 JSON | Multimodal content |
+| 参考图像 | URL 直接传递 | Base64 编码 |
+| 输出格式 | URL | Base64 (需上传) |
+| 价格 | 按 fal.ai 计费 | 根据 Gemini 计费 |
+
+## ⚠️ 注意事项
+
+1. **API 限制**: Google Gemini API 有速率限制,注意不要频繁调用
+2. **图像上传**: 生成的图像需要上传到图床才能发送,建议配置 FAL_KEY
+3. **临时文件**: 脚本会在 `/tmp` 创建临时文件,执行完会自动清理
+4. **OpenClaw**: 如果没有安装 OpenClaw CLI,会尝试直接调用 API
+
+## 🐛 故障排查
+
+### API Key 错误
+```bash
+# 检查 API key 是否设置
+echo $GEMINI_API_KEY
+
+# 测试 API key
+curl -s "https://generativelanguage.googleapis.com/v1beta/models/gemini-3-pro-image-preview:generateContent" \
+  -H "x-goog-api-key: $GEMINI_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d '{"contents":[{"role":"user","parts":[{"text":"test"}]}]}'
+```
+
+### 图像上传失败
+```bash
+# 方案 1: 使用 fal.ai storage
+export FAL_KEY="your_fal_key"
+
+# 方案 2: 检查 imgur 连接
+curl -I https://api.imgur.com/3/image
+
+# 方案 3: 查看本地保存的图像
+ls -lh /tmp/clawra_output_*.png
+```
+
+### OpenClaw 连接失败
+```bash
+# 检查 OpenClaw CLI
+openclaw --version
+
+# 检查 Gateway
+curl http://localhost:18789/health
+
+# 设置自定义 Gateway URL
+export OPENCLAW_GATEWAY_URL="http://your-gateway:port"
+export OPENCLAW_GATEWAY_TOKEN="your_token"
+```
+
+## 📚 相关资源
+
+- [Google AI Studio](https://aistudio.google.com/)
+- [Gemini API 文档](https://ai.google.dev/gemini-api/docs)
+- [OpenClaw 文档](https://openclaw.dev)
+- [Clawra 项目主页](https://github.com/SumeLabs/clawra)
+
+## 🤝 贡献
+
+欢迎提交 Issue 和 Pull Request!
+
+## 📄 许可证
+
+MIT License
diff --git a/scripts/clawra-selfie-with-banana.sh b/scripts/clawra-selfie-with-banana.sh
new file mode 100755
index 0000000..d8750f0
--- /dev/null
+++ b/scripts/clawra-selfie-with-banana.sh
@@ -0,0 +1,308 @@
+#!/bin/bash
+# clawra-selfie-with-banana.sh
+# Generate an image with Google Nano Banana Pro (Gemini 3 Pro Image) and send it via OpenClaw
+#
+# Usage: ./clawra-selfie-with-banana.sh "<prompt>" "<channel>" [options]
+#
+# Environment variables required:
+#   GEMINI_API_KEY - Your Google AI Studio API key
+#
+# Example:
+#   GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh "A sunset over mountains" "#art" "Check this out!"
+
+set -euo pipefail
+
+# Colors for output
+RED='\033[0;31m'
+GREEN='\033[0;32m'
+YELLOW='\033[1;33m'
+CYAN='\033[0;36m'
+NC='\033[0m' # No Color
+
+log_info() {
+    echo -e "${GREEN}[INFO]${NC} $1"
+}
+
+log_warn() {
+    echo -e "${YELLOW}[WARN]${NC} $1"
+}
+
+log_error() {
+    echo -e "${RED}[ERROR]${NC} $1"
+}
+
+log_step() {
+    echo -e "${CYAN}[STEP]${NC} $1"
+}
+
+# Check required environment variables
+if [ -z "${GEMINI_API_KEY:-}" ]; then
+    log_error "GEMINI_API_KEY environment variable not set"
+    echo "Get your API key from: https://aistudio.google.com/apikey"
+    exit 1
+fi
+
+# Check for jq
+if ! command -v jq &> /dev/null; then
+    log_error "jq is required but not installed"
+    echo "Install with: brew install jq (macOS) or apt install jq (Linux)"
+    exit 1
+fi
+
+# Check for openclaw
+if ! command -v openclaw &> /dev/null; then
+    log_warn "openclaw CLI not found - will attempt direct API call"
+    USE_CLI=false
+else
+    USE_CLI=true
+fi
+
+# Parse arguments
+PROMPT="${1:-}"
+CHANNEL="${2:-}"
+CAPTION="${3:-Generated with Nano Banana Pro}"
+MODE="${4:-auto}"  # auto, mirror, direct
+REFERENCE_IMAGE="${5:-}"  # Optional reference image URL
+
+if [ -z "$PROMPT" ] || [ -z "$CHANNEL" ]; then
+    cat << 'EOF'
+Usage: ./clawra-selfie-with-banana.sh <prompt> <channel> [caption] [mode] [reference_image]
+
+Arguments:
+  prompt          - Image description (required)
+  channel         - Target channel (required) e.g., #general, @user
+  caption         - Message caption (default: 'Generated with Nano Banana Pro')
+  mode            - Selfie mode (default: auto) Options: auto, mirror, direct
+  reference_image - Reference image URL for editing (optional)
+
+Modes:
+  auto   - Auto-detect based on keywords in prompt
+  mirror - Full-body mirror selfie (good for outfits)
+  direct - Close-up direct selfie (good for portraits)
+
+Environment:
+  GEMINI_API_KEY  - Your Google AI Studio API key (required)
+                    Get from: https://aistudio.google.com/apikey
+
+Examples:
+  # Text-to-image generation
+  GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+    "A cyberpunk city at night" "#art"
+
+  # Clawra selfie with auto-detection
+  GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+    "wearing a red dress at a party" "#selfies"
+
+  # Direct selfie mode with caption
+  GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+    "at a cozy coffee shop" "#updates" "Morning coffee!" "direct"
+
+  # Image editing with reference
+  GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+    "wearing sunglasses and a hat" "#fun" "New look!" "auto" \
+    "https://example.com/reference.jpg"
+
+EOF
+    exit 1
+fi
+
+# Fixed reference image for Clawra
+CLAWRA_REFERENCE="https://cdn.jsdelivr.net/gh/SumeLabs/clawra@main/assets/clawra.png"
+
+# Determine the reference image to use
+if [ -n "$REFERENCE_IMAGE" ]; then
+    USED_REFERENCE="$REFERENCE_IMAGE"
+    log_info "Using custom reference image: $REFERENCE_IMAGE"
+else
+    USED_REFERENCE="$CLAWRA_REFERENCE"
+    log_info "Using Clawra default reference image"
+fi
+
+# Auto-detect mode based on keywords
+if [ "$MODE" = "auto" ]; then
+    if echo "$PROMPT" | grep -qiE "outfit|wearing|clothes|dress|suit|fashion|full-body|mirror"; then
+        MODE="mirror"
+    elif echo "$PROMPT" | grep -qiE "cafe|restaurant|beach|park|city|close-up|portrait|face|eyes|smile"; then
+        MODE="direct"
+    else
+        MODE="mirror"  # default
+    fi
+    log_info "Auto-detected mode: $MODE"
+fi
+
+# Construct the full prompt based on mode
+if [ "$MODE" = "direct" ]; then
+    FULL_PROMPT="a close-up selfie taken by herself at $PROMPT, direct eye contact with the camera, looking straight into the lens, eyes centered and clearly visible, not a mirror selfie, phone held at arm's length, face fully visible"
+else
+    FULL_PROMPT="make a pic of this person, but $PROMPT. the person is taking a mirror selfie"
+fi
+
+log_step "Generating image with Nano Banana Pro (Gemini 3 Pro Image)..."
+log_info "Mode: $MODE"
+log_info "Prompt: $FULL_PROMPT"
+
+# Prepare the request payload
+# Note: Gemini API uses multimodal content format
+REQUEST_PAYLOAD=$(jq -n \
+    --arg prompt "$FULL_PROMPT" \
+    --arg ref_url "$USED_REFERENCE" \
+    '{
+        "contents": [{
+            "role": "user",
+            "parts": [
+                {
+                    "text": $prompt
+                },
+                {
+                    "inlineData": {
+                        "mimeType": "image/jpeg",
+                        "data": ""
+                    }
+                }
+            ]
+        }],
+        "generationConfig": {
+            "responseModalities": ["IMAGE"],
+            "temperature": 1.0,
+            "candidateCount": 1
+        }
+    }')
+
+# Download reference image and encode to base64
+log_info "Downloading reference image..."
+TEMP_IMAGE="/tmp/clawra_ref_$$.jpg"
+if curl -sL "$USED_REFERENCE" -o "$TEMP_IMAGE"; then
+    BASE64_IMAGE=$(base64 < "$TEMP_IMAGE" | tr -d '\n')
+    rm -f "$TEMP_IMAGE"
+    log_info "Reference image encoded"
+else
+    log_error "Failed to download reference image"
+    exit 1
+fi
+
+# Update payload with base64 image
+REQUEST_PAYLOAD=$(echo "$REQUEST_PAYLOAD" | jq \
+    --arg b64 "$BASE64_IMAGE" \
+    '.contents[0].parts[1].inlineData.data = $b64')
+
+# Generate image via Google Gemini API
+log_step "Calling Gemini API..."
+RESPONSE=$(curl -s -X POST \
+    "https://generativelanguage.googleapis.com/v1beta/models/gemini-3-pro-image-preview:generateContent" \
+    -H "x-goog-api-key: $GEMINI_API_KEY" \
+    -H "Content-Type: application/json" \
+    -d "$REQUEST_PAYLOAD")
+
+# Check for errors in response
+if echo "$RESPONSE" | jq -e '.error' > /dev/null 2>&1; then
+    ERROR_MSG=$(echo "$RESPONSE" | jq -r '.error.message // .error // "Unknown error"')
+    log_error "Image generation failed: $ERROR_MSG"
+    echo "Full response: $RESPONSE"
+    exit 1
+fi
+
+# Extract image data from response
+# Gemini returns base64 image in the response
+IMAGE_DATA=$(echo "$RESPONSE" | jq -r '.candidates[0].content.parts[] | select(.inlineData) | .inlineData.data' 2>/dev/null)
+
+if [ -z "$IMAGE_DATA" ] || [ "$IMAGE_DATA" = "null" ]; then
+    log_error "Failed to extract image data from response"
+    echo "Response: $RESPONSE"
+    exit 1
+fi
+
+log_info "Image generated successfully!"
+
+# Save image temporarily and upload to a hosting service
+# (In production, you might want to use a proper image hosting service)
+TEMP_OUTPUT="/tmp/clawra_output_$$.png"
+echo "$IMAGE_DATA" | base64 -d > "$TEMP_OUTPUT"
+
+log_info "Image saved to: $TEMP_OUTPUT"
+
+# For this example, we'll need to upload the image somewhere
+# Let's use a simple approach with fal.ai storage or imgur
+log_step "Uploading image..."
+
+# Try to upload to fal.ai storage if FAL_KEY is available
+if [ -n "${FAL_KEY:-}" ]; then
+    log_info "Uploading to fal.ai storage..."
+    
+    UPLOAD_RESPONSE=$(curl -s -X POST "https://fal.ai/api/files/upload" \
+        -H "Authorization: Key $FAL_KEY" \
+        -F "file=@$TEMP_OUTPUT")
+    
+    IMAGE_URL=$(echo "$UPLOAD_RESPONSE" | jq -r '.url // empty')
+    
+    if [ -z "$IMAGE_URL" ]; then
+        log_warn "Failed to upload to fal.ai, trying alternative..."
+    fi
+fi
+
+# Fallback: try imgur anonymous upload
+if [ -z "${IMAGE_URL:-}" ]; then
+    log_info "Uploading to imgur..."
+    
+    IMGUR_RESPONSE=$(curl -s -X POST "https://api.imgur.com/3/image" \
+        -H "Authorization: Client-ID 546c25a59c58ad7" \
+        -F "image=@$TEMP_OUTPUT")
+    
+    IMAGE_URL=$(echo "$IMGUR_RESPONSE" | jq -r '.data.link // empty')
+    
+    if [ -z "$IMAGE_URL" ]; then
+        log_error "Failed to upload image to hosting service"
+        log_info "Image saved locally at: $TEMP_OUTPUT"
+        echo "You can manually upload and use it."
+        exit 1
+    fi
+fi
+
+log_info "Image URL: $IMAGE_URL"
+
+# Clean up temp file
+rm -f "$TEMP_OUTPUT"
+
+# Send via OpenClaw
+log_step "Sending to channel: $CHANNEL"
+
+if [ "$USE_CLI" = true ]; then
+    # Use OpenClaw CLI
+    openclaw message send \
+        --action send \
+        --channel "$CHANNEL" \
+        --message "$CAPTION" \
+        --media "$IMAGE_URL"
+else
+    # Direct API call to local gateway
+    GATEWAY_URL="${OPENCLAW_GATEWAY_URL:-http://localhost:18789}"
+    GATEWAY_TOKEN="${OPENCLAW_GATEWAY_TOKEN:-}"
+
+    curl -s -X POST "$GATEWAY_URL/message" \
+        -H "Content-Type: application/json" \
+        ${GATEWAY_TOKEN:+-H "Authorization: Bearer $GATEWAY_TOKEN"} \
+        -d "{
+            \"action\": \"send\",
+            \"channel\": \"$CHANNEL\",
+            \"message\": \"$CAPTION\",
+            \"media\": \"$IMAGE_URL\"
+        }"
+fi
+
+log_info "Done! Image sent to $CHANNEL"
+
+# Output JSON for programmatic use
+echo ""
+echo "--- Result ---"
+jq -n \
+    --arg url "$IMAGE_URL" \
+    --arg channel "$CHANNEL" \
+    --arg prompt "$FULL_PROMPT" \
+    --arg mode "$MODE" \
+    '{
+        success: true,
+        image_url: $url,
+        channel: $channel,
+        prompt: $prompt,
+        mode: $mode,
+        model: "nano-banana-pro"
+    }'
diff --git a/skill/SKILL.md b/skill/SKILL.md
index c09ce1d..16c9735 100644
--- a/skill/SKILL.md
+++ b/skill/SKILL.md
@@ -1,12 +1,17 @@
 ---
 name: clawra-selfie
-description: Edit Clawra's reference image with Grok Imagine (xAI Aurora) and send selfies to messaging channels via OpenClaw
+description: Edit Clawra's reference image with Grok Imagine or Nano Banana Pro and send selfies to messaging channels via OpenClaw
 allowed-tools: Bash(npm:*) Bash(npx:*) Bash(openclaw:*) Bash(curl:*) Read Write WebFetch
 ---
 
 # Clawra Selfie
 
-Edit a fixed reference image using xAI's Grok Imagine model and distribute it across messaging platforms (WhatsApp, Telegram, Discord, Slack, etc.) via OpenClaw.
+Edit a fixed reference image using xAI's Grok Imagine (primary) or Google's Nano Banana Pro (fallback) and distribute it across messaging platforms (WhatsApp, Telegram, Discord, Slack, etc.) via OpenClaw.
+
+## Supported Models
+
+1. **Grok Imagine (Primary)**: xAI Aurora via fal.ai - Fast, high-quality image editing
+2. **Nano Banana Pro (Fallback)**: Google Gemini 3 Pro Image - Used when FAL_KEY is unavailable or fails
 
 ## Reference Image
 
@@ -26,19 +31,33 @@ https://cdn.jsdelivr.net/gh/SumeLabs/clawra@main/assets/clawra.png
 
 ## Quick Reference
 
-### Required Environment Variables
+### Environment Variables (Priority Order)
 
 ```bash
+# Primary: Grok Imagine (xAI via fal.ai)
 FAL_KEY=your_fal_api_key          # Get from https://fal.ai/dashboard/keys
+
+# Fallback: Nano Banana Pro (Google Gemini)
+GEMINI_API_KEY=your_gemini_key    # Get from https://aistudio.google.com/apikey
+
+# OpenClaw (Required)
 OPENCLAW_GATEWAY_TOKEN=your_token  # From: openclaw doctor --generate-gateway-token
 ```
 
+**Fallback Logic:**
+- If `FAL_KEY` is set and valid → Use Grok Imagine
+- If `FAL_KEY` is missing or fails → Use Nano Banana Pro (requires `GEMINI_API_KEY`)
+- If both fail → Return error
+
 ### Workflow
 
 1. **Get user prompt** for how to edit the image
-2. **Edit image** via fal.ai Grok Imagine Edit API with fixed reference
-3. **Extract image URL** from response
-4. **Send to OpenClaw** with target channel(s)
+2. **Choose model**:
+   - Try Grok Imagine (if FAL_KEY available)
+   - Fallback to Nano Banana Pro (if GEMINI_API_KEY available)
+3. **Edit image** via selected model with fixed reference
+4. **Extract/upload image** (Gemini returns base64, needs upload)
+5. **Send to OpenClaw** with target channel(s)
 
 ## Step-by-Step Instructions
 
@@ -85,7 +104,9 @@ a close-up selfie taken by herself at a cozy cafe with warm lighting, direct eye
 | close-up, portrait, face, eyes, smile | `direct` |
 | full-body, mirror, reflection | `mirror` |
 
-### Step 2: Edit Image with Grok Imagine
+### Step 2: Edit Image (Multi-Model Support)
+
+#### Option A: Grok Imagine (Primary)
 
 Use the fal.ai API to edit the reference image:
 
@@ -125,6 +146,81 @@ curl -X POST "https://fal.run/xai/grok-imagine-image/edit" \
 }
 ```
 
+#### Option B: Nano Banana Pro (Fallback)
+
+Use Google Gemini API when fal.ai is unavailable:
+
+```bash
+REFERENCE_IMAGE="https://cdn.jsdelivr.net/gh/SumeLabs/clawra@main/assets/clawra.png"
+
+# Download and encode reference image
+TEMP_IMAGE="/tmp/clawra_ref.jpg"
+curl -sL "$REFERENCE_IMAGE" -o "$TEMP_IMAGE"
+BASE64_IMAGE=$(base64 < "$TEMP_IMAGE" | tr -d '\n')
+
+# Build request payload
+REQUEST_PAYLOAD=$(jq -n \
+  --arg prompt "$PROMPT" \
+  --arg b64 "$BASE64_IMAGE" \
+  '{
+    "contents": [{
+      "role": "user",
+      "parts": [
+        {"text": $prompt},
+        {"inlineData": {"mimeType": "image/jpeg", "data": $b64}}
+      ]
+    }],
+    "generationConfig": {
+      "responseModalities": ["IMAGE"],
+      "temperature": 1.0
+    }
+  }')
+
+# Call Gemini API
+RESPONSE=$(curl -s -X POST \
+  "https://generativelanguage.googleapis.com/v1beta/models/gemini-3-pro-image-preview:generateContent" \
+  -H "x-goog-api-key: $GEMINI_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d "$REQUEST_PAYLOAD")
+
+# Extract base64 image data
+IMAGE_DATA=$(echo "$RESPONSE" | jq -r '.candidates[0].content.parts[] | select(.inlineData) | .inlineData.data')
+
+# Save and upload (since Gemini returns base64)
+echo "$IMAGE_DATA" | base64 -d > "/tmp/clawra_output.png"
+
+# Upload to image hosting (imgur or fal.ai storage)
+if [ -n "${FAL_KEY:-}" ]; then
+  # Upload to fal.ai
+  UPLOAD_RESPONSE=$(curl -s -X POST "https://fal.ai/api/files/upload" \
+    -H "Authorization: Key $FAL_KEY" \
+    -F "file=@/tmp/clawra_output.png")
+  IMAGE_URL=$(echo "$UPLOAD_RESPONSE" | jq -r '.url')
+else
+  # Upload to imgur
+  IMGUR_RESPONSE=$(curl -s -X POST "https://api.imgur.com/3/image" \
+    -H "Authorization: Client-ID 546c25a59c58ad7" \
+    -F "image=@/tmp/clawra_output.png")
+  IMAGE_URL=$(echo "$IMGUR_RESPONSE" | jq -r '.data.link')
+fi
+```
+
+**Response Format:**
+```json
+{
+  "candidates": [{
+    "content": {
+      "parts": [{
+        "inlineData": {
+          "mimeType": "image/png",
+          "data": "base64_encoded_image_data..."
+        }
+      }]
+    }
+  }]
+}
+```
+
 ### Step 3: Send Image via OpenClaw
 
 Use the OpenClaw messaging API to send the edited image:
@@ -150,18 +246,31 @@ curl -X POST "http://localhost:18789/message" \
   }'
 ```
 
-## Complete Script Example
+## Complete Script Example (With Fallback)
 
 ```bash
 #!/bin/bash
-# grok-imagine-edit-send.sh
-
-# Check required environment variables
-if [ -z "$FAL_KEY" ]; then
-  echo "Error: FAL_KEY environment variable not set"
+# clawra-selfie-multi-model.sh
+# Supports both Grok Imagine and Nano Banana Pro
+
+# Check for at least one API key
+if [ -z "${FAL_KEY:-}" ] && [ -z "${GEMINI_API_KEY:-}" ]; then
+  echo "Error: Neither FAL_KEY nor GEMINI_API_KEY is set"
+  echo "Set at least one:"
+  echo "  - FAL_KEY from https://fal.ai/dashboard/keys"
+  echo "  - GEMINI_API_KEY from https://aistudio.google.com/apikey"
   exit 1
 fi
 
+# Determine which model to use
+if [ -n "${FAL_KEY:-}" ]; then
+  MODEL="grok-imagine"
+  echo "Using primary model: Grok Imagine (xAI)"
+else
+  MODEL="nano-banana-pro"
+  echo "Using fallback model: Nano Banana Pro (Google Gemini)"
+fi
+
 # Fixed reference image
 REFERENCE_IMAGE="https://cdn.jsdelivr.net/gh/SumeLabs/clawra@main/assets/clawra.png"
 
@@ -198,26 +307,105 @@ else
 fi
 
 echo "Mode: $MODE"
+echo "Model: $MODEL"
 echo "Editing reference image with prompt: $EDIT_PROMPT"
 
-# Edit image (using jq for proper JSON escaping)
-JSON_PAYLOAD=$(jq -n \
-  --arg image_url "$REFERENCE_IMAGE" \
-  --arg prompt "$EDIT_PROMPT" \
-  '{image_url: $image_url, prompt: $prompt, num_images: 1, output_format: "jpeg"}')
+# Edit image based on selected model
+if [ "$MODEL" == "grok-imagine" ]; then
+  # Grok Imagine via fal.ai
+  JSON_PAYLOAD=$(jq -n \
+    --arg image_url "$REFERENCE_IMAGE" \
+    --arg prompt "$EDIT_PROMPT" \
+    '{image_url: $image_url, prompt: $prompt, num_images: 1, output_format: "jpeg"}')
+
+  RESPONSE=$(curl -s -X POST "https://fal.run/xai/grok-imagine-image/edit" \
+    -H "Authorization: Key $FAL_KEY" \
+    -H "Content-Type: application/json" \
+    -d "$JSON_PAYLOAD")
+
+  # Extract image URL directly
+  IMAGE_URL=$(echo "$RESPONSE" | jq -r '.images[0].url')
+
+  if [ "$IMAGE_URL" == "null" ] || [ -z "$IMAGE_URL" ]; then
+    echo "Error: Grok Imagine failed, trying fallback..."
+    if [ -n "${GEMINI_API_KEY:-}" ]; then
+      MODEL="nano-banana-pro"
+      echo "Switching to Nano Banana Pro"
+    else
+      echo "Response: $RESPONSE"
+      exit 1
+    fi
+  fi
+fi
 
-RESPONSE=$(curl -s -X POST "https://fal.run/xai/grok-imagine-image/edit" \
-  -H "Authorization: Key $FAL_KEY" \
-  -H "Content-Type: application/json" \
-  -d "$JSON_PAYLOAD")
+if [ "$MODEL" == "nano-banana-pro" ]; then
+  # Nano Banana Pro via Google Gemini
+  # Download reference image
+  TEMP_REF="/tmp/clawra_ref_$$.jpg"
+  curl -sL "$REFERENCE_IMAGE" -o "$TEMP_REF"
+  BASE64_IMAGE=$(base64 < "$TEMP_REF" | tr -d '\n')
+  rm -f "$TEMP_REF"
+
+  # Build Gemini request
+  GEMINI_PAYLOAD=$(jq -n \
+    --arg prompt "$EDIT_PROMPT" \
+    --arg b64 "$BASE64_IMAGE" \
+    '{
+      "contents": [{
+        "role": "user",
+        "parts": [
+          {"text": $prompt},
+          {"inlineData": {"mimeType": "image/jpeg", "data": $b64}}
+        ]
+      }],
+      "generationConfig": {
+        "responseModalities": ["IMAGE"],
+        "temperature": 1.0
+      }
+    }')
+
+  RESPONSE=$(curl -s -X POST \
+    "https://generativelanguage.googleapis.com/v1beta/models/gemini-3-pro-image-preview:generateContent" \
+    -H "x-goog-api-key: $GEMINI_API_KEY" \
+    -H "Content-Type: application/json" \
+    -d "$GEMINI_PAYLOAD")
+
+  # Extract base64 image
+  IMAGE_DATA=$(echo "$RESPONSE" | jq -r '.candidates[0].content.parts[] | select(.inlineData) | .inlineData.data')
+
+  if [ -z "$IMAGE_DATA" ] || [ "$IMAGE_DATA" == "null" ]; then
+    echo "Error: Nano Banana Pro failed"
+    echo "Response: $RESPONSE"
+    exit 1
+  fi
 
-# Extract image URL
-IMAGE_URL=$(echo "$RESPONSE" | jq -r '.images[0].url')
+  # Save image
+  TEMP_OUTPUT="/tmp/clawra_output_$$.png"
+  echo "$IMAGE_DATA" | base64 -d > "$TEMP_OUTPUT"
+
+  # Upload to image hosting
+  if [ -n "${FAL_KEY:-}" ]; then
+    echo "Uploading to fal.ai storage..."
+    UPLOAD_RESPONSE=$(curl -s -X POST "https://fal.ai/api/files/upload" \
+      -H "Authorization: Key $FAL_KEY" \
+      -F "file=@$TEMP_OUTPUT")
+    IMAGE_URL=$(echo "$UPLOAD_RESPONSE" | jq -r '.url // empty')
+  fi
 
-if [ "$IMAGE_URL" == "null" ] || [ -z "$IMAGE_URL" ]; then
-  echo "Error: Failed to edit image"
-  echo "Response: $RESPONSE"
-  exit 1
+  if [ -z "${IMAGE_URL:-}" ]; then
+    echo "Uploading to imgur..."
+    IMGUR_RESPONSE=$(curl -s -X POST "https://api.imgur.com/3/image" \
+      -H "Authorization: Client-ID 546c25a59c58ad7" \
+      -F "image=@$TEMP_OUTPUT")
+    IMAGE_URL=$(echo "$IMGUR_RESPONSE" | jq -r '.data.link // empty')
+  fi
+
+  rm -f "$TEMP_OUTPUT"
+
+  if [ -z "$IMAGE_URL" ]; then
+    echo "Error: Failed to upload image"
+    exit 1
+  fi
 fi
 
 echo "Image edited: $IMAGE_URL"
@@ -388,10 +576,24 @@ openclaw gateway start
 
 ## Error Handling
 
-- **FAL_KEY missing**: Ensure the API key is set in environment
-- **Image edit failed**: Check prompt content and API quota
+### API Key Issues
+- **No API keys**: Set either `FAL_KEY` or `GEMINI_API_KEY`
+- **FAL_KEY missing**: Will automatically fallback to Nano Banana Pro
+- **Both keys invalid**: Check key validity and API quotas
+
+### Model-Specific Issues
+- **Grok Imagine failed**: Automatically retries with Nano Banana Pro if `GEMINI_API_KEY` is available
+- **Nano Banana Pro failed**: Check Gemini API quota and rate limits
+- **Image upload failed**: For Gemini, ensure image hosting (imgur/fal.ai) is accessible
+
+### OpenClaw Issues
 - **OpenClaw send failed**: Verify gateway is running and channel exists
-- **Rate limits**: fal.ai has rate limits; implement retry logic if needed
+- **Gateway token missing**: Run `openclaw doctor --generate-gateway-token`
+
+### Rate Limits
+- **fal.ai**: Has rate limits; implement retry logic if needed
+- **Gemini**: Has daily quota limits; monitor usage at https://aistudio.google.com
+- **imgur**: Anonymous uploads have hourly limits
 
 ## Tips
 
diff --git a/skill/scripts/README-BANANA.md b/skill/scripts/README-BANANA.md
new file mode 100644
index 0000000..bf0a13e
--- /dev/null
+++ b/skill/scripts/README-BANANA.md
@@ -0,0 +1,218 @@
+# Clawra Selfie with Nano Banana Pro
+
+使用 Google Gemini (Nano Banana Pro) 生成自拍图像的脚本。
+
+## 🌟 特性
+
+- ✅ 使用 Google Gemini 3 Pro Image Preview API
+- ✅ 支持参考图像编辑 (image-to-image)
+- ✅ 自动模式检测 (mirror/direct selfie)
+- ✅ 集成 OpenClaw 消息发送
+- ✅ 自动上传到图床 (imgur 或 fal.ai)
+- ✅ 彩色日志输出
+- ✅ 完整的错误处理
+
+## 📋 前置要求
+
+### 1. 获取 API Key
+访问 [Google AI Studio](https://aistudio.google.com/apikey) 获取你的 API key。
+
+### 2. 安装依赖
+```bash
+# macOS
+brew install jq curl
+
+# Linux (Debian/Ubuntu)
+apt install jq curl
+
+# OpenClaw (可选,用于发送消息)
+npm install -g openclaw
+```
+
+## 🚀 使用方法
+
+### 基础用法
+
+```bash
+# 设置 API key
+export GEMINI_API_KEY="your_api_key_here"
+
+# 生成图像并发送
+./clawra-selfie-with-banana.sh "prompt" "#channel"
+```
+
+### 完整参数
+
+```bash
+./clawra-selfie-with-banana.sh <prompt> <channel> [caption] [mode] [reference_image]
+```
+
+**参数说明:**
+- `prompt`: 图像描述 (必需)
+- `channel`: 目标频道 (必需), 如 `#general`, `@username`
+- `caption`: 消息标题 (可选, 默认: "Generated with Nano Banana Pro")
+- `mode`: 自拍模式 (可选, 默认: auto)
+  - `auto`: 根据关键词自动检测
+  - `mirror`: 镜子自拍 (全身照)
+  - `direct`: 直接自拍 (特写)
+- `reference_image`: 参考图像 URL (可选, 默认使用 Clawra 官方图像)
+
+## 📝 示例
+
+### 1. 简单文本生成图像
+```bash
+GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+  "A cyberpunk city at night with neon lights" \
+  "#art-gallery"
+```
+
+### 2. Clawra 自拍 (自动模式检测)
+```bash
+# 会自动检测为 mirror 模式 (因为有 "wearing" 关键词)
+GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+  "wearing a red evening dress at a party" \
+  "#selfies" \
+  "Party time! 🎉"
+```
+
+### 3. 指定 direct 模式
+```bash
+# 近景肖像
+GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+  "at a cozy coffee shop with warm lighting" \
+  "#daily-updates" \
+  "Morning coffee ☕" \
+  "direct"
+```
+
+### 4. 使用自定义参考图像
+```bash
+GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+  "wearing sunglasses and a hat" \
+  "#fun" \
+  "New look! 😎" \
+  "auto" \
+  "https://example.com/my-photo.jpg"
+```
+
+### 5. 不发送消息,仅生成图像
+```bash
+# 使用一个无效的 channel,图像会保存在本地
+GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+  "beautiful sunset" \
+  "local"
+```
+
+## 🔧 模式说明
+
+### Mirror 模式
+适用于展示服装、全身照、时尚内容
+
+**触发关键词:**
+- outfit, wearing, clothes, dress, suit, fashion, full-body, mirror
+
+**提示词模板:**
+```
+make a pic of this person, but [你的描述]. the person is taking a mirror selfie
+```
+
+### Direct 模式
+适用于近景肖像、地点拍摄、情感表达
+
+**触发关键词:**
+- cafe, restaurant, beach, park, city, close-up, portrait, face, eyes, smile
+
+**提示词模板:**
+```
+a close-up selfie taken by herself at [你的描述], 
+direct eye contact with the camera, looking straight into the lens, 
+eyes centered and clearly visible, not a mirror selfie, 
+phone held at arm's length, face fully visible
+```
+
+## 🌐 图像上传
+
+脚本会自动尝试上传生成的图像:
+
+1. **优先**: 如果设置了 `FAL_KEY` 环境变量,上传到 fal.ai storage
+2. **备用**: 上传到 imgur 匿名图床
+
+```bash
+# 同时使用两个 API
+export GEMINI_API_KEY="your_gemini_key"
+export FAL_KEY="your_fal_key"  # 可选,用于上传
+
+./clawra-selfie-with-banana.sh "prompt" "#channel"
+```
+
+## 🔄 与原版的区别
+
+| 特性 | Grok Imagine (原版) | Nano Banana Pro (新版) |
+|------|-------------------|---------------------|
+| API 提供商 | xAI (fal.ai) | Google Gemini |
+| API Key | FAL_KEY | GEMINI_API_KEY |
+| 输入格式 | 简单 JSON | Multimodal content |
+| 参考图像 | URL 直接传递 | Base64 编码 |
+| 输出格式 | URL | Base64 (需上传) |
+| 价格 | 按 fal.ai 计费 | 根据 Gemini 计费 |
+
+## ⚠️ 注意事项
+
+1. **API 限制**: Google Gemini API 有速率限制,注意不要频繁调用
+2. **图像上传**: 生成的图像需要上传到图床才能发送,建议配置 FAL_KEY
+3. **临时文件**: 脚本会在 `/tmp` 创建临时文件,执行完会自动清理
+4. **OpenClaw**: 如果没有安装 OpenClaw CLI,会尝试直接调用 API
+
+## 🐛 故障排查
+
+### API Key 错误
+```bash
+# 检查 API key 是否设置
+echo $GEMINI_API_KEY
+
+# 测试 API key
+curl -s "https://generativelanguage.googleapis.com/v1beta/models/gemini-3-pro-image-preview:generateContent" \
+  -H "x-goog-api-key: $GEMINI_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d '{"contents":[{"role":"user","parts":[{"text":"test"}]}]}'
+```
+
+### 图像上传失败
+```bash
+# 方案 1: 使用 fal.ai storage
+export FAL_KEY="your_fal_key"
+
+# 方案 2: 检查 imgur 连接
+curl -I https://api.imgur.com/3/image
+
+# 方案 3: 查看本地保存的图像
+ls -lh /tmp/clawra_output_*.png
+```
+
+### OpenClaw 连接失败
+```bash
+# 检查 OpenClaw CLI
+openclaw --version
+
+# 检查 Gateway
+curl http://localhost:18789/health
+
+# 设置自定义 Gateway URL
+export OPENCLAW_GATEWAY_URL="http://your-gateway:port"
+export OPENCLAW_GATEWAY_TOKEN="your_token"
+```
+
+## 📚 相关资源
+
+- [Google AI Studio](https://aistudio.google.com/)
+- [Gemini API 文档](https://ai.google.dev/gemini-api/docs)
+- [OpenClaw 文档](https://openclaw.dev)
+- [Clawra 项目主页](https://github.com/SumeLabs/clawra)
+
+## 🤝 贡献
+
+欢迎提交 Issue 和 Pull Request!
+
+## 📄 许可证
+
+MIT License
diff --git a/skill/scripts/clawra-selfie-with-banana.sh b/skill/scripts/clawra-selfie-with-banana.sh
new file mode 100755
index 0000000..d8750f0
--- /dev/null
+++ b/skill/scripts/clawra-selfie-with-banana.sh
@@ -0,0 +1,308 @@
+#!/bin/bash
+# clawra-selfie-with-banana.sh
+# Generate an image with Google Nano Banana Pro (Gemini 3 Pro Image) and send it via OpenClaw
+#
+# Usage: ./clawra-selfie-with-banana.sh "<prompt>" "<channel>" [options]
+#
+# Environment variables required:
+#   GEMINI_API_KEY - Your Google AI Studio API key
+#
+# Example:
+#   GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh "A sunset over mountains" "#art" "Check this out!"
+
+set -euo pipefail
+
+# Colors for output
+RED='\033[0;31m'
+GREEN='\033[0;32m'
+YELLOW='\033[1;33m'
+CYAN='\033[0;36m'
+NC='\033[0m' # No Color
+
+log_info() {
+    echo -e "${GREEN}[INFO]${NC} $1"
+}
+
+log_warn() {
+    echo -e "${YELLOW}[WARN]${NC} $1"
+}
+
+log_error() {
+    echo -e "${RED}[ERROR]${NC} $1"
+}
+
+log_step() {
+    echo -e "${CYAN}[STEP]${NC} $1"
+}
+
+# Check required environment variables
+if [ -z "${GEMINI_API_KEY:-}" ]; then
+    log_error "GEMINI_API_KEY environment variable not set"
+    echo "Get your API key from: https://aistudio.google.com/apikey"
+    exit 1
+fi
+
+# Check for jq
+if ! command -v jq &> /dev/null; then
+    log_error "jq is required but not installed"
+    echo "Install with: brew install jq (macOS) or apt install jq (Linux)"
+    exit 1
+fi
+
+# Check for openclaw
+if ! command -v openclaw &> /dev/null; then
+    log_warn "openclaw CLI not found - will attempt direct API call"
+    USE_CLI=false
+else
+    USE_CLI=true
+fi
+
+# Parse arguments
+PROMPT="${1:-}"
+CHANNEL="${2:-}"
+CAPTION="${3:-Generated with Nano Banana Pro}"
+MODE="${4:-auto}"  # auto, mirror, direct
+REFERENCE_IMAGE="${5:-}"  # Optional reference image URL
+
+if [ -z "$PROMPT" ] || [ -z "$CHANNEL" ]; then
+    cat << 'EOF'
+Usage: ./clawra-selfie-with-banana.sh <prompt> <channel> [caption] [mode] [reference_image]
+
+Arguments:
+  prompt          - Image description (required)
+  channel         - Target channel (required) e.g., #general, @user
+  caption         - Message caption (default: 'Generated with Nano Banana Pro')
+  mode            - Selfie mode (default: auto) Options: auto, mirror, direct
+  reference_image - Reference image URL for editing (optional)
+
+Modes:
+  auto   - Auto-detect based on keywords in prompt
+  mirror - Full-body mirror selfie (good for outfits)
+  direct - Close-up direct selfie (good for portraits)
+
+Environment:
+  GEMINI_API_KEY  - Your Google AI Studio API key (required)
+                    Get from: https://aistudio.google.com/apikey
+
+Examples:
+  # Text-to-image generation
+  GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+    "A cyberpunk city at night" "#art"
+
+  # Clawra selfie with auto-detection
+  GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+    "wearing a red dress at a party" "#selfies"
+
+  # Direct selfie mode with caption
+  GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+    "at a cozy coffee shop" "#updates" "Morning coffee!" "direct"
+
+  # Image editing with reference
+  GEMINI_API_KEY=your_key ./clawra-selfie-with-banana.sh \
+    "wearing sunglasses and a hat" "#fun" "New look!" "auto" \
+    "https://example.com/reference.jpg"
+
+EOF
+    exit 1
+fi
+
+# Fixed reference image for Clawra
+CLAWRA_REFERENCE="https://cdn.jsdelivr.net/gh/SumeLabs/clawra@main/assets/clawra.png"
+
+# Determine the reference image to use
+if [ -n "$REFERENCE_IMAGE" ]; then
+    USED_REFERENCE="$REFERENCE_IMAGE"
+    log_info "Using custom reference image: $REFERENCE_IMAGE"
+else
+    USED_REFERENCE="$CLAWRA_REFERENCE"
+    log_info "Using Clawra default reference image"
+fi
+
+# Auto-detect mode based on keywords
+if [ "$MODE" = "auto" ]; then
+    if echo "$PROMPT" | grep -qiE "outfit|wearing|clothes|dress|suit|fashion|full-body|mirror"; then
+        MODE="mirror"
+    elif echo "$PROMPT" | grep -qiE "cafe|restaurant|beach|park|city|close-up|portrait|face|eyes|smile"; then
+        MODE="direct"
+    else
+        MODE="mirror"  # default
+    fi
+    log_info "Auto-detected mode: $MODE"
+fi
+
+# Construct the full prompt based on mode
+if [ "$MODE" = "direct" ]; then
+    FULL_PROMPT="a close-up selfie taken by herself at $PROMPT, direct eye contact with the camera, looking straight into the lens, eyes centered and clearly visible, not a mirror selfie, phone held at arm's length, face fully visible"
+else
+    FULL_PROMPT="make a pic of this person, but $PROMPT. the person is taking a mirror selfie"
+fi
+
+log_step "Generating image with Nano Banana Pro (Gemini 3 Pro Image)..."
+log_info "Mode: $MODE"
+log_info "Prompt: $FULL_PROMPT"
+
+# Prepare the request payload
+# Note: Gemini API uses multimodal content format
+REQUEST_PAYLOAD=$(jq -n \
+    --arg prompt "$FULL_PROMPT" \
+    --arg ref_url "$USED_REFERENCE" \
+    '{
+        "contents": [{
+            "role": "user",
+            "parts": [
+                {
+                    "text": $prompt
+                },
+                {
+                    "inlineData": {
+                        "mimeType": "image/jpeg",
+                        "data": ""
+                    }
+                }
+            ]
+        }],
+        "generationConfig": {
+            "responseModalities": ["IMAGE"],
+            "temperature": 1.0,
+            "candidateCount": 1
+        }
+    }')
+
+# Download reference image and encode to base64
+log_info "Downloading reference image..."
+TEMP_IMAGE="/tmp/clawra_ref_$$.jpg"
+if curl -sL "$USED_REFERENCE" -o "$TEMP_IMAGE"; then
+    BASE64_IMAGE=$(base64 < "$TEMP_IMAGE" | tr -d '\n')
+    rm -f "$TEMP_IMAGE"
+    log_info "Reference image encoded"
+else
+    log_error "Failed to download reference image"
+    exit 1
+fi
+
+# Update payload with base64 image
+REQUEST_PAYLOAD=$(echo "$REQUEST_PAYLOAD" | jq \
+    --arg b64 "$BASE64_IMAGE" \
+    '.contents[0].parts[1].inlineData.data = $b64')
+
+# Generate image via Google Gemini API
+log_step "Calling Gemini API..."
+RESPONSE=$(curl -s -X POST \
+    "https://generativelanguage.googleapis.com/v1beta/models/gemini-3-pro-image-preview:generateContent" \
+    -H "x-goog-api-key: $GEMINI_API_KEY" \
+    -H "Content-Type: application/json" \
+    -d "$REQUEST_PAYLOAD")
+
+# Check for errors in response
+if echo "$RESPONSE" | jq -e '.error' > /dev/null 2>&1; then
+    ERROR_MSG=$(echo "$RESPONSE" | jq -r '.error.message // .error // "Unknown error"')
+    log_error "Image generation failed: $ERROR_MSG"
+    echo "Full response: $RESPONSE"
+    exit 1
+fi
+
+# Extract image data from response
+# Gemini returns base64 image in the response
+IMAGE_DATA=$(echo "$RESPONSE" | jq -r '.candidates[0].content.parts[] | select(.inlineData) | .inlineData.data' 2>/dev/null)
+
+if [ -z "$IMAGE_DATA" ] || [ "$IMAGE_DATA" = "null" ]; then
+    log_error "Failed to extract image data from response"
+    echo "Response: $RESPONSE"
+    exit 1
+fi
+
+log_info "Image generated successfully!"
+
+# Save image temporarily and upload to a hosting service
+# (In production, you might want to use a proper image hosting service)
+TEMP_OUTPUT="/tmp/clawra_output_$$.png"
+echo "$IMAGE_DATA" | base64 -d > "$TEMP_OUTPUT"
+
+log_info "Image saved to: $TEMP_OUTPUT"
+
+# For this example, we'll need to upload the image somewhere
+# Let's use a simple approach with fal.ai storage or imgur
+log_step "Uploading image..."
+
+# Try to upload to fal.ai storage if FAL_KEY is available
+if [ -n "${FAL_KEY:-}" ]; then
+    log_info "Uploading to fal.ai storage..."
+    
+    UPLOAD_RESPONSE=$(curl -s -X POST "https://fal.ai/api/files/upload" \
+        -H "Authorization: Key $FAL_KEY" \
+        -F "file=@$TEMP_OUTPUT")
+    
+    IMAGE_URL=$(echo "$UPLOAD_RESPONSE" | jq -r '.url // empty')
+    
+    if [ -z "$IMAGE_URL" ]; then
+        log_warn "Failed to upload to fal.ai, trying alternative..."
+    fi
+fi
+
+# Fallback: try imgur anonymous upload
+if [ -z "${IMAGE_URL:-}" ]; then
+    log_info "Uploading to imgur..."
+    
+    IMGUR_RESPONSE=$(curl -s -X POST "https://api.imgur.com/3/image" \
+        -H "Authorization: Client-ID 546c25a59c58ad7" \
+        -F "image=@$TEMP_OUTPUT")
+    
+    IMAGE_URL=$(echo "$IMGUR_RESPONSE" | jq -r '.data.link // empty')
+    
+    if [ -z "$IMAGE_URL" ]; then
+        log_error "Failed to upload image to hosting service"
+        log_info "Image saved locally at: $TEMP_OUTPUT"
+        echo "You can manually upload and use it."
+        exit 1
+    fi
+fi
+
+log_info "Image URL: $IMAGE_URL"
+
+# Clean up temp file
+rm -f "$TEMP_OUTPUT"
+
+# Send via OpenClaw
+log_step "Sending to channel: $CHANNEL"
+
+if [ "$USE_CLI" = true ]; then
+    # Use OpenClaw CLI
+    openclaw message send \
+        --action send \
+        --channel "$CHANNEL" \
+        --message "$CAPTION" \
+        --media "$IMAGE_URL"
+else
+    # Direct API call to local gateway
+    GATEWAY_URL="${OPENCLAW_GATEWAY_URL:-http://localhost:18789}"
+    GATEWAY_TOKEN="${OPENCLAW_GATEWAY_TOKEN:-}"
+
+    curl -s -X POST "$GATEWAY_URL/message" \
+        -H "Content-Type: application/json" \
+        ${GATEWAY_TOKEN:+-H "Authorization: Bearer $GATEWAY_TOKEN"} \
+        -d "{
+            \"action\": \"send\",
+            \"channel\": \"$CHANNEL\",
+            \"message\": \"$CAPTION\",
+            \"media\": \"$IMAGE_URL\"
+        }"
+fi
+
+log_info "Done! Image sent to $CHANNEL"
+
+# Output JSON for programmatic use
+echo ""
+echo "--- Result ---"
+jq -n \
+    --arg url "$IMAGE_URL" \
+    --arg channel "$CHANNEL" \
+    --arg prompt "$FULL_PROMPT" \
+    --arg mode "$MODE" \
+    '{
+        success: true,
+        image_url: $url,
+        channel: $channel,
+        prompt: $prompt,
+        mode: $mode,
+        model: "nano-banana-pro"
+    }'