@baimingtao/lbs-agent 1.0.0

package/extensions/skills/optional/creative/meme-generation/EXAMPLES.md

@@ -0,0 +1,46 @@
+# Meme Generation 示例
+
+## 示例 1: 凌晨 2 点 Debugging
+
+**主题：** 凌晨 2 点 debugging 生产环境
+**模板：** this-is-fine
+
+```bash
+python generate_meme.py this-is-fine /tmp/meme.png "PRODUCTION IS DOWN" "This is fine"
+```
+
+## 示例 2: 开发者优先级
+
+**主题：** 在编写测试与发布功能之间做选择
+**模板：** drake
+
+```bash
+python generate_meme.py drake /tmp/meme.png "Writing unit tests" "Shipping straight to prod"
+```
+
+## 示例 3: 考试压力
+
+**主题：** 期末考试准备
+**模板：** two-buttons
+
+```bash
+python generate_meme.py two-buttons /tmp/meme.png "Study everything" "Sleep" "Me at midnight"
+```
+
+## 示例 4: 解决方案不断升级
+
+**主题：** 修复 CSS bug
+**模板：** expanding-brain
+
+```bash
+python generate_meme.py expanding-brain /tmp/meme.png "Reading the docs" "Stack Overflow" "!important on everything" "Deleting the stylesheet"
+```
+
+## 示例 5: 犀利观点
+
+**主题：** tabs 与 spaces 之争
+**模板：** change-my-mind
+
+```bash
+python generate_meme.py change-my-mind /tmp/meme.png "Tabs are just thicc spaces"
+```

package/extensions/skills/optional/creative/meme-generation/SKILL.md

@@ -0,0 +1,130 @@
+---
+name: meme-generation
+description: 通过选择模板并使用 Pillow 叠加文字来生成真实的 meme 图像。生成实际的 .png meme 文件。
+version: 2.0.0
+author: adanaleycio
+license: MIT
+platforms: [linux, macos, windows]
+metadata:
+  hermes:
+    tags: [creative, memes, humor, images]
+    related_skills: [ascii-art, generative-widgets]
+    category: creative
+---
+
+# Meme 生成
+
+根据主题生成真实的 meme 图像。选择模板，编写标题，并渲染带有文本叠加的真实 .png 文件。
+
+## 何时使用
+
+- 用户要求你制作或生成 meme
+- 用户想要关于特定主题、情境或挫败感的 meme
+- 用户说“做个 meme”（meme this）或类似的话
+
+## 可用模板
+
+该脚本支持通过名称或 ID 选择约 100 种流行的 imgflip 模板中的任意一种，外加 10 种带有手动调整文本位置的精选模板。
+
+### 精选模板（自定义文本位置）
+
+| ID | 名称 | 字段 | 最适用于 |
+| ---- | ------ | -------- | ---------- |
+| `this-is-fine` | This is Fine | top, bottom | 混乱、否认 |
+| `drake` | Drake Hotline Bling | reject, approve | 拒绝/偏好 |
+| `distracted-boyfriend` | Distracted Boyfriend | distraction, current, person | 诱惑、转移注意力 |
+| `two-buttons` | Two Buttons | left, right, person | 无法做出的选择 |
+| `expanding-brain` | Expanding Brain | 4 levels | 不断升级的讽刺 |
+| `change-my-mind` | Change My Mind | statement | 尖锐观点 |
+| `woman-yelling-at-cat` | Woman Yelling at Cat | woman, cat | 争论 |
+| `one-does-not-simply` | One Does Not Simply | top, bottom | 看似简单实则极难的事 |
+| `grus-plan` | Gru's Plan | step1-3, realization | 适得其反的计划 |
+| `batman-slapping-robin` | Batman Slapping Robin | robin, batman | 制止糟糕的想法 |
+
+### 动态模板（来自 imgflip API）
+
+任何不在精选列表中的模板都可以通过名称或 imgflip ID 使用。这些模板将获得智能默认文本位置（2个字段为顶部/底部，3个及以上为均匀间隔）。搜索方式：
+```bash
+python "$SKILL_DIR/scripts/generate_meme.py" --search "disaster"
+```
+
+## 流程
+
+### 模式 1：经典模板（默认）
+
+1. 阅读用户的主题并识别核心动态（混乱、困境、偏好、讽刺等）。
+2. 挑选最匹配的模板。使用“最适用于”列，或使用 `--search` 进行搜索。
+3. 为每个字段编写简短的标题（每个字段最多 8-12 个单词，越短越好）。
+4. 找到该技能的脚本目录：
+   ```
+   SKILL_DIR=$(dirname "$(find ~/.hermes/skills -path '*/meme-generation/SKILL.md' 2>/dev/null | head -1)")
+   ```
+5. 运行生成器：
+   ```bash
+   python "$SKILL_DIR/scripts/generate_meme.py" <template_id> /tmp/meme.png "caption 1" "caption 2" ...
+   ```
+6. 使用 `MEDIA:/tmp/meme.png` 返回图像。
+
+### 模式 2：自定义 AI 图像（当 image_generate 可用时）
+
+当没有经典模板匹配，或者用户想要一些原创内容时使用此模式。
+
+1. 首先编写标题。
+2. 使用 `image_generate` 创建与 meme 概念相匹配的场景。请勿在图像提示词中包含任何文本——文本将由脚本添加。仅描述视觉场景。
+3. 从 image_generate 结果 URL 中找到生成的图像路径。如有必要，将其下载到本地路径。
+4. 运行带有 `--image` 的脚本以叠加文本，并选择一种模式：
+   - **Overlay**（文本直接显示在图像上，白色字体带黑色描边）：
+     ```bash
+     python "$SKILL_DIR/scripts/generate_meme.py" --image /path/to/scene.png /tmp/meme.png "top text" "bottom text"
+     ```
+   - **Bars**（图像上方/下方有黑条，内含白色文本——更整洁，且始终易于阅读）：
+     ```bash
+     python "$SKILL_DIR/scripts/generate_meme.py" --image /path/to/scene.png --bars /tmp/meme.png "top text" "bottom text"
+     ```
+   当图像元素较多/细节丰富且在上方难以阅读文本时，请使用 `--bars`。
+5. **使用视觉功能验证**（如果 `vision_analyze` 可用）：检查结果是否理想：
+   ```
+   vision_analyze(image_url="/tmp/meme.png", question="Is the text legible and well-positioned? Does the meme work visually?")
+   ```
+   如果视觉模型指出问题（文本难以阅读、位置不佳等），请尝试另一种模式（在 overlay 和 bars 之间切换）或重新生成场景。
+6. 使用 `MEDIA:/tmp/meme.png` 返回图像。
+
+## 示例
+
+**“凌晨 2 点调试生产环境”：**
+```bash
+python generate_meme.py this-is-fine /tmp/meme.png "SERVERS ARE ON FIRE" "This is fine"
+```
+
+**“在睡觉和多看一集之间做选择”：**
+```bash
+python generate_meme.py drake /tmp/meme.png "Getting 8 hours of sleep" "One more episode at 3 AM"
+```
+
+**“星期一早晨的各个阶段”：**
+```bash
+python generate_meme.py expanding-brain /tmp/meme.png "Setting an alarm" "Setting 5 alarms" "Sleeping through all alarms" "Working from bed"
+```
+
+## 列出模板
+
+查看所有可用模板：
+```bash
+python generate_meme.py --list
+```
+
+## 注意事项
+
+- 保持标题简短。带有长文本的 meme 效果很差。
+- 使文本参数的数量与模板的字段数相匹配。
+- 选择适合笑话结构的模板，而不仅仅是契合主题。
+- 请勿生成仇恨、虐待或针对个人的内容。
+- 脚本在首次下载后会将模板图像缓存在 `scripts/.cache/` 中。
+
+## 验证
+
+如果满足以下条件，则输出正确：
+- 已在输出路径创建了 .png 文件
+- 模板上的文本清晰可辨（白色字体带黑色描边）
+- 笑点成立——标题与模板预期的结构相匹配
+- 文件可以通过 MEDIA: 路径交付

package/extensions/skills/optional/creative/meme-generation/scripts/generate_meme.py

@@ -0,0 +1,470 @@
+#!/usr/bin/env python3
+"""Generate a meme image by overlaying text on a template.
+
+Usage:
+    python generate_meme.py <template_id_or_name> <output_path> <text1> [text2] [text3] [text4]
+
+Example:
+    python generate_meme.py drake /tmp/meme.png "Writing tests" "Shipping to prod and hoping"
+    python generate_meme.py "Disaster Girl" /tmp/meme.png "Top text" "Bottom text"
+    python generate_meme.py --list                    # 显示精选模板
+    python generate_meme.py --search "distracted"     # 搜索所有 imgflip 模板
+
+Templates with custom text positioning are in templates.json (10 curated).
+Any of the ~100 popular imgflip templates can also be used by name or ID —
+unknown templates get smart default text positioning based on their box_count.
+"""
+
+import json
+import os
+import sys
+from io import BytesIO
+from pathlib import Path
+
+try:
+    import requests as _requests
+except ImportError:
+    _requests = None
+
+from PIL import Image, ImageDraw, ImageFont
+
+SCRIPT_DIR = Path(__file__).parent
+TEMPLATES_FILE = SCRIPT_DIR / "templates.json"
+CACHE_DIR = SCRIPT_DIR / ".cache"
+IMGFLIP_API = "https://api.imgflip.com/get_memes"
+IMGFLIP_CACHE_FILE = CACHE_DIR / "imgflip_memes.json"
+IMGFLIP_CACHE_MAX_AGE = 86400  # 24小时
+
+
+def _fetch_url(url: str, timeout: int = 15) -> bytes:
+    """Fetch URL content, using requests if available, else urllib."""
+    if _requests is not None:
+        resp = _requests.get(url, timeout=timeout)
+        resp.raise_for_status()
+        return resp.content
+    import urllib.request
+    return urllib.request.urlopen(url, timeout=timeout).read()
+
+
+def load_curated_templates() -> dict:
+    """Load templates with hand-tuned text field positions."""
+    with open(TEMPLATES_FILE) as f:
+        return json.load(f)
+
+
+def _default_fields(box_count: int) -> list:
+    """Generate sensible default text field positions for unknown templates."""
+    if box_count <= 0:
+        box_count = 2
+    if box_count == 1:
+        return [{"name": "text", "x_pct": 0.5, "y_pct": 0.5, "w_pct": 0.90, "align": "center"}]
+    if box_count == 2:
+        return [
+            {"name": "top", "x_pct": 0.5, "y_pct": 0.08, "w_pct": 0.95, "align": "center"},
+            {"name": "bottom", "x_pct": 0.5, "y_pct": 0.92, "w_pct": 0.95, "align": "center"},
+        ]
+    # 3个及以上：垂直均匀分布
+    fields = []
+    for i in range(box_count):
+        y = 0.08 + (0.84 * i / (box_count - 1)) if box_count > 1 else 0.5
+        fields.append({
+            "name": f"text{i+1}",
+            "x_pct": 0.5,
+            "y_pct": round(y, 2),
+            "w_pct": 0.90,
+            "align": "center",
+        })
+    return fields
+
+
+def fetch_imgflip_templates() -> list:
+    """Fetch popular meme templates from imgflip API. Cached for 24h."""
+    import time
+
+    CACHE_DIR.mkdir(exist_ok=True)
+    # 检查缓存
+    if IMGFLIP_CACHE_FILE.exists():
+        age = time.time() - IMGFLIP_CACHE_FILE.stat().st_mtime
+        if age < IMGFLIP_CACHE_MAX_AGE:
+            with open(IMGFLIP_CACHE_FILE) as f:
+                return json.load(f)
+
+    try:
+        data = json.loads(_fetch_url(IMGFLIP_API))
+        memes = data.get("data", {}).get("memes", [])
+        with open(IMGFLIP_CACHE_FILE, "w") as f:
+            json.dump(memes, f)
+        return memes
+    except Exception as e:
+        # 如果获取失败且存在过期缓存，则使用过期缓存
+        if IMGFLIP_CACHE_FILE.exists():
+            with open(IMGFLIP_CACHE_FILE) as f:
+                return json.load(f)
+        print(f"Warning: could not fetch imgflip templates: {e}", file=sys.stderr)
+        return []
+
+
+def _slugify(name: str) -> str:
+    """Convert a template name to a slug for matching."""
+    return name.lower().replace(" ", "-").replace("'", "").replace("\"", "")
+
+
+def resolve_template(identifier: str) -> dict:
+    """Resolve a template by curated ID, imgflip name, or imgflip ID.
+
+    Returns dict with: name, url, fields, source.
+    """
+    curated = load_curated_templates()
+
+    # 1. 精确匹配精选 ID
+    if identifier in curated:
+        tmpl = curated[identifier]
+        return {**tmpl, "source": "curated"}
+
+    # 2. 转换为 Slug 格式的精选匹配
+    slug = _slugify(identifier)
+    for tid, tmpl in curated.items():
+        if _slugify(tmpl["name"]) == slug or tid == slug:
+            return {**tmpl, "source": "curated"}
+
+    # 3. 搜索 imgflip 模板
+    imgflip_memes = fetch_imgflip_templates()
+    slug_lower = slug.lower()
+    id_lower = identifier.strip()
+
+    for meme in imgflip_memes:
+        meme_slug = _slugify(meme["name"])
+        # 为此 imgflip 模板优先检查精选（自定义定位）
+        for tid, ctmpl in curated.items():
+            if _slugify(ctmpl["name"]) == meme_slug:
+                if meme_slug == slug_lower or meme["id"] == id_lower:
+                    return {**ctmpl, "source": "curated"}
+
+        if meme_slug == slug_lower or meme["id"] == id_lower or slug_lower in meme_slug:
+            return {
+                "name": meme["name"],
+                "url": meme["url"],
+                "fields": _default_fields(meme.get("box_count", 2)),
+                "source": "imgflip",
+            }
+
+    return None
+
+
+def get_template_image(url: str) -> Image.Image:
+    """Download a template image, caching it locally."""
+    CACHE_DIR.mkdir(exist_ok=True)
+    # 使用 URL 哈希作为缓存键
+    cache_name = url.split("/")[-1]
+    cache_path = CACHE_DIR / cache_name
+
+    # 始终缓存为 PNG 格式以避免 JPEG/RGBA 冲突
+    cache_path = cache_path.with_suffix(".png")
+
+    if cache_path.exists():
+        return Image.open(cache_path).convert("RGBA")
+
+    data = _fetch_url(url)
+    img = Image.open(BytesIO(data)).convert("RGBA")
+    img.save(cache_path, "PNG")
+    return img
+
+
+def find_font(size: int) -> ImageFont.FreeTypeFont:
+    """Find a bold font for meme text. Tries Impact, then falls back."""
+    candidates = [
+        "/usr/share/fonts/truetype/msttcorefonts/Impact.ttf",
+        "/usr/share/fonts/truetype/liberation/LiberationSans-Bold.ttf",
+        "/usr/share/fonts/liberation-sans/LiberationSans-Bold.ttf",
+        "/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf",
+        "/usr/share/fonts/dejavu-sans/DejaVuSans-Bold.ttf",
+        "/System/Library/Fonts/Helvetica.ttc",
+        "/System/Library/Fonts/SFCompact.ttf",
+    ]
+    for path in candidates:
+        if os.path.exists(path):
+            try:
+                return ImageFont.truetype(path, size)
+            except (OSError, IOError):
+                continue
+    # 最后手段：Pillow 默认值
+    try:
+        return ImageFont.truetype("DejaVuSans-Bold", size)
+    except (OSError, IOError):
+        return ImageFont.load_default()
+
+
+def _wrap_text(text: str, font: ImageFont.FreeTypeFont, max_width: int) -> str:
+    """Word-wrap text to fit within max_width pixels. Never breaks mid-word."""
+    words = text.split()
+    if not words:
+        return text
+    lines = []
+    current_line = words[0]
+    for word in words[1:]:
+        test_line = current_line + " " + word
+        if font.getlength(test_line) <= max_width:
+            current_line = test_line
+        else:
+            lines.append(current_line)
+            current_line = word
+    lines.append(current_line)
+    return "\n".join(lines)
+
+
+def draw_outlined_text(
+    draw: ImageDraw.ImageDraw,
+    text: str,
+    x: int,
+    y: int,
+    font_size: int,
+    max_width: int,
+    align: str = "center",
+):
+    """Draw white text with black outline, auto-scaled to fit max_width."""
+    # 自动缩放：减小字体大小直到文本合理适应
+    size = font_size
+    while size > 12:
+        font = find_font(size)
+        wrapped = _wrap_text(text, font, max_width)
+        bbox = draw.multiline_textbbox((0, 0), wrapped, font=font, align=align)
+        text_w = bbox[2] - bbox[0]
+        line_count = wrapped.count("\n") + 1
+        # 如果宽度合适且行数不多则接受
+        if text_w <= max_width * 1.05 and line_count <= 4:
+            break
+        size -= 2
+    else:
+        font = find_font(size)
+        wrapped = _wrap_text(text, font, max_width)
+
+    # 测量整个文本块
+    bbox = draw.multiline_textbbox((0, 0), wrapped, font=font, align=align)
+    text_w = bbox[2] - bbox[0]
+    text_h = bbox[3] - bbox[1]
+
+    # 水平居中于 x，垂直居中于 y
+    tx = x - text_w // 2
+    ty = y - text_h // 2
+
+    # 绘制轮廓（黑色边框）
+    outline_range = max(2, font.size // 18)
+    for dx in range(-outline_range, outline_range + 1):
+        for dy in range(-outline_range, outline_range + 1):
+            if dx == 0 and dy == 0:
+                continue
+            draw.multiline_text(
+                (tx + dx, ty + dy), wrapped, font=font, fill="black", align=align
+            )
+    # 绘制主要文本（白色）
+    draw.multiline_text((tx, ty), wrapped, font=font, fill="white", align=align)
+
+
+def _overlay_on_image(img: Image.Image, texts: list, fields: list) -> Image.Image:
+    """Overlay meme text directly on an image using field positions."""
+    draw = ImageDraw.Draw(img)
+    w, h = img.size
+    base_font_size = max(16, min(w, h) // 12)
+
+    for i, field in enumerate(fields):
+        if i >= len(texts):
+            break
+        text = texts[i].strip()
+        if not text:
+            continue
+        fx = int(field["x_pct"] * w)
+        fy = int(field["y_pct"] * h)
+        fw = int(field["w_pct"] * w)
+        draw_outlined_text(draw, text, fx, fy, base_font_size, fw, field.get("align", "center"))
+    return img
+
+
+def _add_bars(img: Image.Image, texts: list) -> Image.Image:
+    """Add black bars with white text above/below the image.
+
+    Distributes texts across bars: first text on top bar, last text on
+    bottom bar, any middle texts overlaid on the image center.
+    """
+    w, h = img.size
+    bar_font_size = max(20, w // 16)
+    font = find_font(bar_font_size)
+    padding = bar_font_size // 2
+
+    top_text = texts[0].strip() if texts else ""
+    bottom_text = texts[-1].strip() if len(texts) > 1 else ""
+    middle_texts = [t.strip() for t in texts[1:-1]] if len(texts) > 2 else []
+
+    def _measure_bar(text: str) -> int:
+        if not text:
+            return 0
+        wrapped = _wrap_text(text, font, int(w * 0.92))
+        bbox = ImageDraw.Draw(Image.new("RGB", (1, 1))).multiline_textbbox(
+            (0, 0), wrapped, font=font, align="center"
+        )
+        return (bbox[3] - bbox[1]) + padding * 2
+
+    top_h = _measure_bar(top_text)
+    bottom_h = _measure_bar(bottom_text)
+    new_h = h + top_h + bottom_h
+
+    canvas = Image.new("RGB", (w, new_h), (0, 0, 0))
+    canvas.paste(img.convert("RGB"), (0, top_h))
+    draw = ImageDraw.Draw(canvas)
+
+    if top_text:
+        wrapped = _wrap_text(top_text, font, int(w * 0.92))
+        bbox = draw.multiline_textbbox((0, 0), wrapped, font=font, align="center")
+        tw = bbox[2] - bbox[0]
+        th = bbox[3] - bbox[1]
+        tx = (w - tw) // 2
+        ty = (top_h - th) // 2
+        draw.multiline_text((tx, ty), wrapped, font=font, fill="white", align="center")
+
+    if bottom_text:
+        wrapped = _wrap_text(bottom_text, font, int(w * 0.92))
+        bbox = draw.multiline_textbbox((0, 0), wrapped, font=font, align="center")
+        tw = bbox[2] - bbox[0]
+        th = bbox[3] - bbox[1]
+        tx = (w - tw) // 2
+        ty = top_h + h + (bottom_h - th) // 2
+        draw.multiline_text((tx, ty), wrapped, font=font, fill="white", align="center")
+
+    # 将任意中间文本居中覆盖在图片上
+    if middle_texts:
+        mid_fields = _default_fields(len(middle_texts))
+        # 调整 y 坐标以抵消顶部栏的偏移
+        for field in mid_fields:
+            field["y_pct"] = (top_h + field["y_pct"] * h) / new_h
+            field["w_pct"] = 0.90
+        _overlay_on_image(canvas, middle_texts, mid_fields)
+
+    return canvas
+
+
+def generate_meme(template_id: str, texts: list[str], output_path: str) -> str:
+    """Generate a meme from a template and save it. Returns the path."""
+    tmpl = resolve_template(template_id)
+
+    if tmpl is None:
+        print(f"Unknown template: {template_id}", file=sys.stderr)
+        print("Use --list to see curated templates or --search to find imgflip templates.", file=sys.stderr)
+        sys.exit(1)
+
+    fields = tmpl["fields"]
+    print(f"Using template: {tmpl['name']} ({tmpl['source']}, {len(fields)} fields)", file=sys.stderr)
+
+    img = get_template_image(tmpl["url"])
+    img = _overlay_on_image(img, texts, fields)
+
+    output = Path(output_path)
+    if output.suffix.lower() in {".jpg", ".jpeg"}:
+        img = img.convert("RGB")
+    img.save(str(output), quality=95)
+    return str(output)
+
+
+def generate_from_image(
+    image_path: str, texts: list[str], output_path: str, use_bars: bool = False
+) -> str:
+    """Generate a meme from a custom image (e.g. AI-generated). Returns the path."""
+    img = Image.open(image_path).convert("RGBA")
+    print(f"Custom image: {img.size[0]}x{img.size[1]}, {len(texts)} text(s), mode={'bars' if use_bars else 'overlay'}", file=sys.stderr)
+
+    if use_bars:
+        result = _add_bars(img, texts)
+    else:
+        fields = _default_fields(len(texts))
+        result = _overlay_on_image(img, texts, fields)
+
+    output = Path(output_path)
+    if output.suffix.lower() in {".jpg", ".jpeg"}:
+        result = result.convert("RGB")
+    result.save(str(output), quality=95)
+    return str(output)
+
+
+def list_templates():
+    """Print curated templates with custom positioning."""
+    templates = load_curated_templates()
+    print(f"{'ID':<25} {'Name':<30} {'Fields':<8} Best for")
+    print("-" * 90)
+    for tid, tmpl in sorted(templates.items()):
+        fields = len(tmpl["fields"])
+        print(f"{tid:<25} {tmpl['name']:<30} {fields:<8} {tmpl['best_for']}")
+    print(f"\n{len(templates)} curated templates with custom text positioning.")
+    print("Use --search to find any of the ~100 popular imgflip templates.")
+
+
+def search_templates(query: str):
+    """Search imgflip templates by name."""
+    imgflip_memes = fetch_imgflip_templates()
+    curated = load_curated_templates()
+    curated_slugs = {_slugify(t["name"]) for t in curated.values()}
+    query_lower = query.lower()
+
+    matches = []
+    for meme in imgflip_memes:
+        if query_lower in meme["name"].lower():
+            slug = _slugify(meme["name"])
+            has_custom = "curated" if slug in curated_slugs else "default"
+            matches.append((meme["name"], meme["id"], meme.get("box_count", 2), has_custom))
+
+    if not matches:
+        print(f"No templates found matching '{query}'")
+        return
+
+    print(f"{'Name':<40} {'ID':<12} {'Fields':<8} Positioning")
+    print("-" * 75)
+    for name, mid, boxes, positioning in matches:
+        print(f"{name:<40} {mid:<12} {boxes:<8} {positioning}")
+    print(f"\n{len(matches)} template(s) found. Use the name or ID as the first argument.")
+
+
+if __name__ == "__main__":
+    if len(sys.argv) < 2:
+        print("Usage: generate_meme.py <template_id_or_name> <output_path> <text1> [text2] ...")
+        print("       generate_meme.py --image <path> [--bars] <output_path> <text1> [text2] ...")
+        print("       generate_meme.py --list              # curated templates")
+        print("       generate_meme.py --search <query>    # search all imgflip templates")
+        sys.exit(1)
+
+    if sys.argv[1] == "--list":
+        list_templates()
+        sys.exit(0)
+
+    if sys.argv[1] == "--search":
+        if len(sys.argv) < 3:
+            print("Usage: generate_meme.py --search <query>")
+            sys.exit(1)
+        search_templates(sys.argv[2])
+        sys.exit(0)
+
+    if sys.argv[1] == "--image":
+        # 自定义图片模式：--image <path> [--bars] <output> <text1> ...
+        args = sys.argv[2:]
+        if len(args) < 3:
+            print("Usage: generate_meme.py --image <image_path> [--bars] <output_path> <text1> ...")
+            sys.exit(1)
+        image_path = args.pop(0)
+        use_bars = False
+        if args and args[0] == "--bars":
+            use_bars = True
+            args.pop(0)
+        if len(args) < 2:
+            print("Need at least: output_path and one text argument")
+            sys.exit(1)
+        output_path = args.pop(0)
+        result = generate_from_image(image_path, args, output_path, use_bars=use_bars)
+        print(f"Meme saved to: {result}")
+        sys.exit(0)
+
+    if len(sys.argv) < 4:
+        print("Need at least: template_id_or_name, output_path, and one text argument")
+        sys.exit(1)
+
+    template_id = sys.argv[1]
+    output_path = sys.argv[2]
+    texts = sys.argv[3:]
+
+    result = generate_meme(template_id, texts, output_path)
+    print(f"Meme saved to: {result}")