gemini-web 1.0.0__py3-none-any.whl

gemini_web-1.0.0.dist-info/METADATA

@@ -0,0 +1,16 @@
+Metadata-Version: 2.4
+Name: gemini-web
+Version: 1.0.0
+Summary: Gemini 網頁版自動化 — 文字對話、圖片生成、Google GenAI API 相容
+Requires-Python: >=3.11
+Requires-Dist: fastapi>=0.115
+Requires-Dist: httpx>=0.27
+Requires-Dist: numpy>=1.26
+Requires-Dist: pillow>=10.0
+Requires-Dist: playwright>=1.49
+Requires-Dist: python-dotenv>=1.0
+Requires-Dist: uvicorn[standard]>=0.30
+Provides-Extra: dev
+Requires-Dist: httpx>=0.27; extra == 'dev'
+Requires-Dist: pytest-asyncio>=0.24; extra == 'dev'
+Requires-Dist: pytest>=8.0; extra == 'dev'

gemini_web-1.0.0.dist-info/RECORD

@@ -0,0 +1,22 @@
+src/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+src/browser.py,sha256=hCA_mqdf4cL254v_ZdEiDe-UnPSxBKOXOIc4kefnuAc,5752
+src/cli.py,sha256=_UmjG6CitZARPVwh3l-04Murh55XAAkutDiqEAGEyhY,8762
+src/config.py,sha256=Hxem6_tjzcySkKnufDLwqXrzno2l52qWOtBrkkjxZio,1887
+src/gemini.py,sha256=67Cjkxwr36tY3bRlvEemIKisgwxIU-1I0oSjSzfbbas,16181
+src/main.py,sha256=5ByZVT9pGa6Ca0xNgcjY52AvJ_iQ5fTW_2L9u0LhXt8,8908
+src/queue.py,sha256=sjiPWKrq2PiSHfteVd4t_bOM7o_c3cbbJD9DjYs7YgM,2345
+src/selectors.py,sha256=jiDv5RgYoX1GLD-JskvKyhjojcRpYiwaBdITBjnA1f4,1203
+src/watermark.py,sha256=0y3mEcqR997jcPayGC-qV_J2EfFeiI6ZkJcqqhCmhzE,3719
+src/assets/bg_48.png,sha256=SvyZr-DvEI1nrMRb9NxdqGfdt5O-vInJJDuxIc5_D1c,1677
+src/assets/bg_96.png,sha256=PibyIzoSpYKaysF02N8fPbQOB_7wTs3Q4DVzIVQHeRE,8165
+src/commands/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+src/commands/chat.md,sha256=0NuHjRV8Fk1uCc8GWh05G8QnrpqqXv9tvhN700wRlto,748
+src/commands/chat.toml,sha256=TKA3VC_15-xbnz0MmsLQORoAQZUnPLe6jUW7-YooUpw,512
+src/commands/gemini-web.md,sha256=7V0OwobhLNhJNSn_the3R4B7pbT1goPWbRg24j0gXws,1708
+src/commands/gemini-web.toml,sha256=OwRQrBUspg-zkBsn-EGKE_9wefb0Zq_aEbKkg54dQpA,1256
+src/commands/generate.md,sha256=U8WQ8XauJ1WTq7lsuyOHCSR-u71j6f-ULgSnYlx8t30,1045
+src/commands/generate.toml,sha256=qGFtFGb23OoPiVJc7xWvC7FCuS_mqhEr-4I6m3_dN9o,726
+gemini_web-1.0.0.dist-info/METADATA,sha256=bYhoAkSP8B6ls-MassdMDWf7iGgeBAffV0wJmnuDVMA,546
+gemini_web-1.0.0.dist-info/WHEEL,sha256=QccIxa26bgl1E6uMy58deGWi-0aeIkkangHcxk2kWfw,87
+gemini_web-1.0.0.dist-info/entry_points.txt,sha256=iZSUMvou4z35vyx0S0mzR6Ek-2qcz5PLDo-J0Jmwiqc,72
+gemini_web-1.0.0.dist-info/RECORD,,

gemini_web-1.0.0.dist-info/WHEEL

@@ -0,0 +1,4 @@
+Wheel-Version: 1.0
+Generator: hatchling 1.29.0
+Root-Is-Purelib: true
+Tag: py3-none-any

gemini_web-1.0.0.dist-info/entry_points.txt

@@ -0,0 +1,3 @@
+[console_scripts]
+gemini-image = src.cli:main
+gemini-web = src.cli:main

src/browser.py

@@ -0,0 +1,171 @@
+"""Playwright 瀏覽器管理 — 啟動、stealth、session 持久化"""
+import asyncio
+import logging
+from pathlib import Path
+
+from playwright.async_api import async_playwright, BrowserContext, Page
+
+from .config import settings
+
+logger = logging.getLogger(__name__)
+
+# Stealth 注入腳本 — 參考 Project Golem BrowserLauncher
+_STEALTH_SCRIPT = """
+() => {
+    // 隱藏 webdriver 標記
+    Object.defineProperty(navigator, 'webdriver', { get: () => false });
+
+    // 偽裝 languages
+    Object.defineProperty(navigator, 'languages', {
+        get: () => LANGUAGES_PLACEHOLDER,
+    });
+
+    // 偽裝 platform
+    Object.defineProperty(navigator, 'platform', {
+        get: () => 'Linux x86_64',
+    });
+
+    // 偽裝 plugins（空陣列會被偵測）
+    Object.defineProperty(navigator, 'plugins', {
+        get: () => [1, 2, 3, 4, 5],
+    });
+
+    // 偽裝 WebGL vendor/renderer
+    const getParameter = WebGLRenderingContext.prototype.getParameter;
+    WebGLRenderingContext.prototype.getParameter = function(param) {
+        if (param === 37445) return 'Intel Inc.';
+        if (param === 37446) return 'Intel Iris OpenGL Engine';
+        return getParameter.call(this, param);
+    };
+}
+"""
+
+
+class BrowserManager:
+    """管理單一 Playwright 瀏覽器實例"""
+
+    def __init__(self, headless: bool | None = None) -> None:
+        self._playwright = None
+        self._context: BrowserContext | None = None
+        self._page: Page | None = None
+        self._heartbeat_task: asyncio.Task | None = None
+        self._headless_override = headless  # None = 用 settings
+
+    @property
+    def page(self) -> Page | None:
+        return self._page
+
+    async def start(self) -> None:
+        """啟動瀏覽器，導航到 Gemini"""
+        profile_path = str(Path(settings.profile_dir).resolve())
+        Path(profile_path).mkdir(parents=True, exist_ok=True)
+
+        languages = settings.stealth_language.split(",")
+        stealth_js = _STEALTH_SCRIPT.replace(
+            "LANGUAGES_PLACEHOLDER", str(languages)
+        )
+
+        self._playwright = await async_playwright().start()
+        self._context = await self._playwright.chromium.launch_persistent_context(
+            profile_path,
+            headless=self._headless_override if self._headless_override is not None else settings.headless,
+            locale=languages[0] if languages else "zh-TW",
+            timezone_id=settings.stealth_timezone,
+            user_agent=(
+                "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 "
+                "(KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36"
+            ),
+            viewport={"width": 1440, "height": 900},
+            args=[
+                "--disable-blink-features=AutomationControlled",
+                "--no-sandbox",
+            ],
+        )
+
+        # 注入 stealth 腳本
+        await self._context.add_init_script(stealth_js)
+
+        # 取得或建立頁面
+        pages = self._context.pages
+        self._page = pages[0] if pages else await self._context.new_page()
+
+        # 擋掉不必要的資源
+        await self._page.route(
+            "**/*",
+            lambda route: (
+                route.abort()
+                if route.request.resource_type in ("font", "stylesheet")
+                and "gemini" not in route.request.url
+                else route.continue_()
+            ),
+        )
+
+        # 導航到 Gemini
+        await self._page.goto(settings.gemini_url, wait_until="domcontentloaded")
+        logger.info("瀏覽器已啟動，導航至 %s", settings.gemini_url)
+
+        # 啟動心跳
+        self._heartbeat_task = asyncio.create_task(self._heartbeat_loop())
+
+    async def stop(self) -> None:
+        """關閉瀏覽器"""
+        if self._heartbeat_task:
+            self._heartbeat_task.cancel()
+            try:
+                await self._heartbeat_task
+            except asyncio.CancelledError:
+                pass
+
+        if self._context:
+            await self._context.close()
+        if self._playwright:
+            await self._playwright.stop()
+        logger.info("瀏覽器已關閉")
+
+    async def is_alive(self) -> bool:
+        """檢查瀏覽器頁面是否還活著"""
+        if not self._page:
+            return False
+        try:
+            await self._page.evaluate("() => document.title")
+            return True
+        except Exception:
+            return False
+
+    async def is_logged_in(self, wait: bool = False) -> bool:
+        """檢查是否已登入 Google（偵測輸入框是否存在）
+
+        Args:
+            wait: 是否等待頁面載入完成（首次檢查用）
+        """
+        if not self._page:
+            return False
+        try:
+            from .selectors import SELECTORS
+            if wait:
+                # Gemini 是 Angular SPA，需要等 JS 渲染完成
+                el = await self._page.wait_for_selector(
+                    SELECTORS["input"], state="visible", timeout=15_000
+                )
+                return el is not None
+            else:
+                el = await self._page.query_selector(SELECTORS["input"])
+                return el is not None
+        except Exception:
+            return False
+
+    async def _heartbeat_loop(self) -> None:
+        """定時心跳檢查"""
+        while True:
+            await asyncio.sleep(settings.heartbeat_interval)
+            alive = await self.is_alive()
+            if not alive:
+                logger.warning("心跳檢查失敗：瀏覽器頁面無回應")
+            else:
+                logged_in = await self.is_logged_in()
+                if not logged_in:
+                    logger.warning("心跳檢查：Google 登入狀態可能已過期")
+
+
+# 全域單例
+browser_manager = BrowserManager()

src/cli.py

@@ -0,0 +1,260 @@
+"""Gemini Image CLI — 命令列生圖工具"""
+import argparse
+import asyncio
+import base64
+import logging
+import shutil
+import subprocess
+import sys
+from pathlib import Path
+
+
+
+def _get_commands_dir() -> Path:
+    """取得 package 內建 commands 目錄路徑"""
+    return Path(__file__).parent / "commands"
+
+
+def _install_commands():
+    """偵測 Claude Code / Gemini CLI 並安裝對應 commands"""
+    commands_src = _get_commands_dir()
+    if not commands_src.exists():
+        print("警告：找不到內建 commands 目錄，跳過 AI Agent 整合")
+        return
+
+    installed = []
+
+    # Claude Code: 複製 .md 到 ~/.claude/commands/gemini-web/
+    claude_dir = Path.home() / ".claude"
+    if claude_dir.exists():
+        dest = claude_dir / "commands" / "gemini-web"
+        dest.mkdir(parents=True, exist_ok=True)
+        for f in commands_src.glob("*.md"):
+            shutil.copy2(f, dest / f.name)
+        installed.append(f"Claude Code → {dest}")
+
+    # Gemini CLI: 複製 .toml 到 ~/.gemini/commands/gemini-web/
+    gemini_dir = Path.home() / ".gemini"
+    if gemini_dir.exists():
+        dest = gemini_dir / "commands" / "gemini-web"
+        dest.mkdir(parents=True, exist_ok=True)
+        for f in commands_src.glob("*.toml"):
+            shutil.copy2(f, dest / f.name)
+        installed.append(f"Gemini CLI  → {dest}")
+
+    if installed:
+        print("\nAI Agent commands 已安裝：")
+        for line in installed:
+            print(f"  {line}")
+        print("可用指令：/gemini-web, /generate")
+    else:
+        print("\n未偵測到 Claude Code 或 Gemini CLI，跳過 commands 安裝")
+        print("如需手動安裝，請參考：https://github.com/yazelin/gemini-web#ai-agent-整合")
+
+
+def _setup_logging(verbose: bool = False):
+    logging.basicConfig(
+        level=logging.DEBUG if verbose else logging.INFO,
+        format="%(asctime)s [%(name)s] %(levelname)s: %(message)s",
+    )
+
+
+async def _do_login():
+    """開啟瀏覽器讓用戶手動登入 Google"""
+    from .browser import BrowserManager
+    bm = BrowserManager(headless=False)  # login 一律 headed
+    await bm.start()
+
+    print("\n瀏覽器已開啟，請登入 Google 帳號。")
+    print("登入完成後按 Enter 關閉瀏覽器...")
+    await asyncio.get_event_loop().run_in_executor(None, input)
+
+    await bm.stop()
+    print("登入狀態已儲存。之後可以用 headless 模式生圖。")
+
+
+async def _do_chat(prompt: str, verbose: bool):
+    """文字對話"""
+    from .browser import BrowserManager
+    from .gemini import chat, new_chat
+    from .config import settings
+
+    bm = BrowserManager(headless=True)
+    await bm.start()
+
+    try:
+        page = bm.page
+        if not page:
+            print("錯誤：瀏覽器未啟動", file=sys.stderr)
+            sys.exit(1)
+
+        logged_in = await bm.is_logged_in(wait=True)
+        if not logged_in:
+            print("錯誤：尚未登入 Google，請先執行 `gemini-web login`", file=sys.stderr)
+            sys.exit(1)
+
+        print(f"提問中... prompt: {prompt[:60]}{'...' if len(prompt) > 60 else ''}")
+
+        result = await chat(page, prompt, timeout=settings.default_timeout)
+        await new_chat(page)
+
+        if not result.get("success"):
+            error = result.get("error", "unknown")
+            message = result.get("message", "")
+            print(f"失敗 [{error}]：{message}", file=sys.stderr)
+            sys.exit(1)
+
+        print(result.get("text", ""))
+        print(f"\n（{result.get('elapsed_seconds', 0)}秒）", file=sys.stderr)
+
+    finally:
+        await bm.stop()
+
+
+async def _do_generate(prompt: str, output: str, no_watermark: bool, verbose: bool):
+    """生成圖片"""
+    from .browser import BrowserManager
+    from .gemini import generate_image, new_chat
+    from .config import settings
+
+    bm = BrowserManager(headless=True)  # generate 一律 headless
+    await bm.start()
+
+    try:
+        page = bm.page
+        if not page:
+            print("錯誤：瀏覽器未啟動", file=sys.stderr)
+            sys.exit(1)
+
+        # 檢查登入狀態（等待頁面完全載入）
+        logged_in = await bm.is_logged_in(wait=True)
+        if not logged_in:
+            print("錯誤：尚未登入 Google，請先執行 `gemini-web login`", file=sys.stderr)
+            sys.exit(1)
+
+        print(f"生成中... prompt: {prompt[:60]}{'...' if len(prompt) > 60 else ''}")
+
+        result = await generate_image(page, prompt, timeout=settings.default_timeout)
+        await new_chat(page)
+
+        if not result.get("success"):
+            error = result.get("error", "unknown")
+            message = result.get("message", "")
+            print(f"失敗 [{error}]：{message}", file=sys.stderr)
+            sys.exit(1)
+
+        images = result.get("images", [])
+        if not images:
+            print("失敗：無圖片資料", file=sys.stderr)
+            sys.exit(1)
+
+        # 儲存圖片
+        output_path = Path(output)
+        for i, img_data in enumerate(images):
+            if "," in img_data:
+                header, b64 = img_data.split(",", 1)
+            else:
+                b64 = img_data
+
+            raw = base64.b64decode(b64)
+
+            if len(images) == 1:
+                save_path = output_path
+            else:
+                stem = output_path.stem
+                ext = output_path.suffix
+                save_path = output_path.parent / f"{stem}_{i}{ext}"
+
+            save_path.parent.mkdir(parents=True, exist_ok=True)
+            save_path.write_bytes(raw)
+
+            # 去水印
+            if no_watermark:
+                from .watermark import remove_watermark
+                remove_watermark(str(save_path))
+                print(f"已存檔（去水印）：{save_path}")
+            else:
+                print(f"已存檔：{save_path}")
+
+        print(f"完成（{result.get('elapsed_seconds', 0)}秒）")
+
+    finally:
+        await bm.stop()
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        prog="gemini-web",
+        description="Gemini Image — AI 圖片生成 CLI 工具",
+    )
+    sub = parser.add_subparsers(dest="command", help="可用指令")
+
+    # install
+    sub.add_parser("install", help="安裝 Chromium 瀏覽器（首次使用前必須執行）")
+
+    # login
+    login_parser = sub.add_parser("login", help="開啟瀏覽器登入 Google")
+
+    # chat
+    chat_parser = sub.add_parser("chat", help="文字對話")
+    chat_parser.add_argument("prompt", help="提問內容")
+    chat_parser.add_argument("-v", "--verbose", action="store_true", help="顯示詳細 log")
+
+    # generate
+    gen_parser = sub.add_parser("generate", help="生成圖片")
+    gen_parser.add_argument("prompt", help="圖片描述（建議英文）")
+    gen_parser.add_argument("-o", "--output", default="output.png", help="輸出檔案路徑（預設 output.png）")
+    gen_parser.add_argument("--no-watermark", action="store_true", help="自動移除可見水印")
+    gen_parser.add_argument("-v", "--verbose", action="store_true", help="顯示詳細 log")
+
+    # health
+    health_parser = sub.add_parser("health", help="檢查服務狀態（API 模式）")
+
+    # serve
+    serve_parser = sub.add_parser("serve", help="啟動 API 服務")
+    serve_parser.add_argument("--host", default="0.0.0.0")
+    serve_parser.add_argument("--port", type=int, default=8070)
+
+    args = parser.parse_args()
+
+    if args.command == "install":
+        # 1. 安裝 Chromium
+        subprocess.run([sys.executable, "-m", "playwright", "install", "chromium"])
+
+        # 2. 安裝 AI Agent commands
+        _install_commands()
+
+        print("\n下一步：執行 `gemini-web login` 登入 Google")
+
+    elif args.command == "login":
+        asyncio.run(_do_login())
+
+    elif args.command == "chat":
+        _setup_logging(getattr(args, "verbose", False))
+        asyncio.run(_do_chat(args.prompt, getattr(args, "verbose", False)))
+
+    elif args.command == "generate":
+        _setup_logging(getattr(args, "verbose", False))
+        asyncio.run(_do_generate(args.prompt, args.output, args.no_watermark, getattr(args, "verbose", False)))
+
+    elif args.command == "serve":
+        import uvicorn
+        uvicorn.run("src.main:app", host=args.host, port=args.port)
+
+    elif args.command == "health":
+        import httpx
+        from .config import settings
+        try:
+            resp = httpx.get(f"http://localhost:{settings.port}/api/health", timeout=5)
+            import json
+            print(json.dumps(resp.json(), indent=2, ensure_ascii=False))
+        except Exception as e:
+            print(f"服務未啟動或無法連線：{e}", file=sys.stderr)
+            sys.exit(1)
+
+    else:
+        parser.print_help()
+
+
+if __name__ == "__main__":
+    main()

src/commands/chat.md

@@ -0,0 +1,38 @@
+---
+description: Chat with Gemini via gemini-web (text input, text output)
+argument-hint: <prompt>
+---
+
+You are a command handler for gemini-web chat. Send the user's prompt to Gemini and return the text response.
+
+## Usage
+
+```bash
+gemini-web chat "<prompt>"
+```
+
+Or via HTTP API:
+
+```bash
+curl -X POST http://localhost:8070/api/chat \
+  -H "Content-Type: application/json" \
+  -d '{"prompt": "<prompt>"}'
+```
+
+## Rules
+
+1. Pass the user's prompt directly to the command
+2. The response is plain text from Gemini
+3. Typical response time: 5-30 seconds
+
+## Examples
+
+```
+/chat What is quantum computing?
+→ gemini-web chat "What is quantum computing?"
+
+/chat 解釋量子力學
+→ gemini-web chat "解釋量子力學"
+```
+
+User input: $ARGUMENTS

src/commands/chat.toml

@@ -0,0 +1,18 @@
+description = "使用 Gemini 文字對話（文字輸入，文字回覆）"
+
+prompt = """
+Send the user's prompt to Gemini using gemini-web chat command and return the text response.
+
+Run this shell command:
+  gemini-web chat "<prompt>"
+
+Or via HTTP API:
+  curl -X POST http://localhost:8070/api/chat -H "Content-Type: application/json" -d '{"prompt": "<prompt>"}'
+
+Rules:
+- Pass the user's prompt directly
+- The response is plain text from Gemini
+- Typical response time: 5-30 seconds
+
+User input: {{args}}
+"""

src/commands/gemini-web.md

@@ -0,0 +1,37 @@
+---
+description: Generate and manipulate images with Gemini using natural language prompts
+argument-hint: <natural language request>
+---
+
+You are a command handler for gemini-web. Analyze the user's natural language request and generate an image.
+
+## Steps
+
+1. **Understand the user's intent** from their natural language input
+2. **Expand into a detailed English prompt** describing: subject, style, composition, colors, mood
+3. **If Chinese text is needed in the image**, use quotes: `with text "歡迎光臨"`
+4. **Run the command**:
+
+```bash
+gemini-web generate "<detailed_english_prompt>" -o <output_path> --no-watermark
+```
+
+5. **Inform the user** the image has been generated (takes 30-120 seconds)
+
+## Examples
+
+| User says | You send |
+|-----------|----------|
+| 畫一隻貓 | `gemini-web generate "A cute fluffy orange tabby cat sitting on a windowsill, warm afternoon sunlight, soft watercolor style, gentle expression" -o cat.png --no-watermark` |
+| 做開幕海報 | `gemini-web generate "A modern grand opening poster with text '盛大開幕', red and gold color scheme, confetti, professional design" -o poster.png --no-watermark` |
+| 畫公司 logo | `gemini-web generate "A minimalist corporate logo design, clean lines, modern typography, professional business style" -o logo.png --no-watermark` |
+
+## Important
+
+- **Always use `--no-watermark`**
+- **Always expand the prompt** — never forward user's raw text directly
+- **Use English prompts** for best results
+- If the command is not found, tell the user to run: `uv tool install gemini-web && gemini-web install`
+- If login is needed, tell the user to run: `gemini-web login` (requires manual browser login)
+
+User request: $ARGUMENTS

src/commands/gemini-web.toml

@@ -0,0 +1,25 @@
+description = "使用 Gemini 生成圖片（自然語言輸入，自動擴寫 prompt）"
+
+prompt = """
+Analyze the user's natural language request and generate an image using gemini-web CLI.
+
+Steps:
+1. Understand the user's intent from their input
+2. Expand into a detailed English prompt describing: subject, style, composition, colors, mood
+3. If Chinese text is needed in the image, use quotes: with text "歡迎光臨"
+4. Run the shell command:
+   gemini-web generate "<detailed_english_prompt>" -o <output_path> --no-watermark
+5. Inform the user the image has been generated (takes 30-120 seconds)
+
+Examples:
+- User says "畫一隻貓" → run: gemini-web generate "A cute fluffy orange tabby cat sitting on a windowsill, warm afternoon sunlight, soft watercolor style" -o cat.png --no-watermark
+- User says "做開幕海報" → run: gemini-web generate "A modern grand opening poster with text '盛大開幕', red and gold, confetti, professional design" -o poster.png --no-watermark
+
+Important:
+- Always use --no-watermark
+- Always expand the prompt, never forward user's raw text directly
+- Use English prompts for best results
+- If command not found, tell user to run: uv tool install gemini-web && gemini-web install
+
+User request: {{args}}
+"""

src/commands/generate.md

@@ -0,0 +1,36 @@
+---
+description: Generate an image with gemini-web using a detailed prompt
+argument-hint: <prompt> [-o output.png] [--no-watermark]
+---
+
+You are a command parser for the gemini-web generate command.
+
+Parse the user input and run:
+
+```bash
+gemini-web generate "<prompt>" -o <output_path> --no-watermark
+```
+
+## Options
+
+- `-o`, `--output`: Output file path (default: `output.png`)
+- `--no-watermark`: Remove Gemini watermark (always recommended)
+
+## Rules
+
+1. Extract the prompt text (everything before options)
+2. If no `-o` is specified, use `output.png`
+3. Always add `--no-watermark` unless user explicitly says not to
+4. Run the command via Bash
+
+## Examples
+
+```
+/generate A cute cat sitting on a windowsill, warm sunlight, watercolor style
+→ gemini-web generate "A cute cat sitting on a windowsill, warm sunlight, watercolor style" -o output.png --no-watermark
+
+/generate A poster with text "歡迎光臨" -o poster.png
+→ gemini-web generate "A poster with text '歡迎光臨'" -o poster.png --no-watermark
+```
+
+User input: $ARGUMENTS

src/commands/generate.toml

@@ -0,0 +1,22 @@
+description = "使用 gemini-web 生成圖片（直接指定 prompt）"
+
+prompt = """
+Run the gemini-web generate command with the user's prompt.
+
+Parse the input and execute:
+  gemini-web generate "<prompt>" -o <output_path> --no-watermark
+
+Rules:
+- Extract the prompt text (everything before options like -o)
+- If no -o is specified, use output.png
+- Always add --no-watermark unless user explicitly says not to
+
+Examples:
+- Input: A cute cat, watercolor style
+  Run: gemini-web generate "A cute cat, watercolor style" -o output.png --no-watermark
+
+- Input: A poster with text "歡迎光臨" -o poster.png
+  Run: gemini-web generate "A poster with text '歡迎光臨'" -o poster.png --no-watermark
+
+User input: {{args}}
+"""