Back to Details

anycap-social-meme-workflows

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

AnyCap Social Meme Workflows

AnyCap 社交表情包工作流

Use this skill when the output needs to feel like a meme, captioned social post, or reaction visual, but still has to be reproducible.
Do not rely on image generation alone for exact caption text. Use AnyCap to create or edit the base visual, then render the final caption locally so the text is exact.
当输出内容需要具备表情包、带字幕社交帖或反应类视觉内容的风格，同时又要求可复现性时，可使用本技能。
请勿仅依赖图像生成来实现精确的字幕文本。 使用AnyCap创建或编辑基础视觉素材，然后通过本地渲染生成最终字幕，确保文本完全准确。

Read First

必读内容

Read these files before acting:
  1. This file for the workflow
  2. references/workflows.md for pattern selection, prompt formulas, and article mapping
For detailed CLI syntax, authentication, and capability reference, use the 
anycap-cli
skill.
执行操作前请先阅读以下文件：
  1. 本文件（工作流说明）
  2. references/workflows.md（用于模式选择、提示公式和文章映射）
如需详细的CLI语法、认证和功能参考，请使用
anycap-cli
技能。

Best Fit

适用场景

Use this skill for:
  • meme-style hero images with exact top or bottom text
  • funny meme drawings with doodle-style internet humor
  • captioned photos for blog posts or social posts
  • reaction visuals from an existing screenshot or photo
  • short meme-video concepts where the still or base frame comes first
  • use-case demos that need both generated media and a repeatable workflow
Do not use this skill for:
  • large meme-template databases
  • highly specific internet meme lore pages
  • exact brand or copyrighted character recreation requests
  • production subtitle pipelines with timing-heavy caption editing
本技能适用于：
  • 带有精确顶部或底部文本的表情包风格主图
  • 带有涂鸦式网络幽默的搞笑手绘表情包
  • 用于博客文章或社交帖的带字幕照片
  • 基于现有截图或照片制作的反应类视觉内容
  • 先有静态图或基础帧的短视频表情包概念
  • 需要同时生成媒体内容和可重复工作流的用例演示
本技能不适用于：
  • 大型表情包模板数据库
  • 高度特定的网络表情包知识库页面
  • 精确复刻品牌或受版权保护角色的请求
  • 对字幕编辑时序要求严格的生产级字幕流水线

Core Rule

核心规则

Split the task into two layers:
  1. Base visual layer with AnyCap
  2. Exact text layer with deterministic local rendering
Why:
  • image models are good at style, composition, and fast variation
  • image models are not dependable for long exact caption text
  • deterministic overlay keeps the final meme readable and repeatable
将任务拆分为两个层级：
  1. 基础视觉层：使用AnyCap生成或编辑
  2. 精确文本层：通过确定性本地渲染实现
原因：
  • 图像模型擅长风格、构图和快速变体生成
  • 图像模型无法可靠生成长段精确字幕文本
  • 确定性文本叠加可保证最终表情包的可读性和可复现性

Workflow

工作流

mermaid
graph LR
    A[Classify request] --> B[Choose model]
    B --> C[Generate or edit base visual with AnyCap]
    C --> D[Overlay exact text locally]
    D --> E[QA readability and punchline]
    E --> F[Deliver locally, via Drive, or via Page]
mermaid
graph LR
    A[Classify request] --> B[Choose model]
    B --> C[Generate or edit base visual with AnyCap]
    C --> D[Overlay exact text locally]
    D --> E[QA readability and punchline]
    E --> F[Deliver locally, via Drive, or via Page]

1. Classify the request

1. 分类请求

Choose one workflow first:
  • Text-first meme: the joke or caption exists; the visual supports it
  • Funny meme drawing: the humor mostly lives in the drawing style, pose, or absurd scene
  • Reaction remix: user supplies an image and wants meme treatment
  • Captioned photo: exact line of text on top of an image
  • Meme-video concept: still image, caption, then optional short video
For funny meme drawings, default to one of these repeatable presets:
  • Classic doodle: the strongest default for blob characters, stick-figure-adjacent humor, and easy-to-draw meme pages
  • Bad drawing: useful when the joke works because the art is awkward or deliberately clumsy
  • Rage-comic-adjacent: only when you want old-web comic energy without relying on canonical rage faces or meme-lore cloning
首先选择一种工作流：
  • 文本优先型表情包：已有笑话或字幕，视觉内容为其提供支撑
  • 搞笑手绘表情包：幽默主要体现在绘画风格、姿势或荒诞场景中
  • 反应类重制：用户提供图片，需要进行表情包化处理
  • 带字幕照片：在图片上添加精确文本
  • 短视频表情包概念：先制作静态图和字幕，再按需生成短视频
对于搞笑手绘表情包，默认使用以下可重复预设之一：
  • 经典涂鸦：最适合 blob 角色、类简笔画幽默和易绘制的表情包页面
  • 粗糙画风：当笑话的笑点在于艺术表现笨拙或刻意粗糙时适用
  • 类暴走漫画风格：仅当需要旧网络漫画风格，但又不想依赖标准暴走表情或表情包知识库克隆时使用

2. Choose the model

2. 选择模型

Choose from the live model catalog instead of relying on a hard-coded model list. Match the model to the job:
  • stronger first-pass visuals
  • editing an existing image or screenshot
  • fast variant generation, especially for funny meme drawings
Always inspect the model list and schema before generating:
bash
anycap image models
anycap image models <model-id> schema --operation generate --mode <mode>
从实时模型目录中选择，而非依赖硬编码模型列表。根据任务匹配模型：
  • 首次生成视觉效果更出色
  • 编辑现有图像或截图
  • 快速生成变体，尤其适用于搞笑手绘表情包
生成前务必检查模型列表和 schema：
bash
anycap image models
anycap image models <model-id> schema --operation generate --mode <mode>

3. Generate or edit the base visual

3. 生成或编辑基础视觉素材

Text-to-image example:
bash
anycap image generate \
  --model <model-id> \
  --prompt "reaction-image style visual, exaggerated expression, blank top and bottom safe space for meme caption, high contrast, clean composition" \
  --param aspect_ratio=1:1 \
  --param resolution=2k \
  -o meme-base.png
Image-to-image example:
bash
anycap image generate \
  --model <model-id> \
  --mode image-to-image \
  --prompt "turn this into a sharper reaction meme image, preserve the subject, simplify background, leave clear safe space for top and bottom caption" \
  --param images=./source.png \
  --param aspect_ratio=1:1 \
  --param resolution=2k \
  -o meme-remix-base.png
Prompt for negative space explicitly. Ask for "blank caption-safe area", "clean top band", or "empty bottom margin" instead of asking the model to write the exact meme text.
Funny meme drawings example:
bash
anycap image generate \
  --model <model-id> \
  --prompt "funny meme drawing, crude but charming internet doodle style, exhausted office goblin melting into an office chair while holding a tiny coffee cup, absurd tiny-problem energy, wildly exaggerated defeated expression, messy desk chaos without readable screens, thick sketch lines, off-white paper texture, muted green accents, obvious blank space for optional caption, no words, no letters, no watermark" \
  --param aspect_ratio=4:3 \
  --param resolution=2k \
  -o funny-meme-drawing.png
For funny meme drawings, the caption is optional. If the humor already lands through the drawing alone, you can deliver the image as-is. If the joke needs exact wording, add the caption locally afterward.
文本转图像示例：
bash
anycap image generate \
  --model <model-id> \
  --prompt "reaction-image style visual, exaggerated expression, blank top and bottom safe space for meme caption, high contrast, clean composition" \
  --param aspect_ratio=1:1 \
  --param resolution=2k \
  -o meme-base.png
图像转图像示例：
bash
anycap image generate \
  --model <model-id> \
  --mode image-to-image \
  --prompt "turn this into a sharper reaction meme image, preserve the subject, simplify background, leave clear safe space for top and bottom caption" \
  --param images=./source.png \
  --param aspect_ratio=1:1 \
  --param resolution=2k \
  -o meme-remix-base.png
明确要求预留空白区域。使用“caption-safe空白区域”“顶部干净区域”或“底部空白边距”等描述，而非要求模型直接生成精确的表情包文本。
搞笑手绘表情包示例：
bash
anycap image generate \
  --model <model-id> \
  --prompt "funny meme drawing, crude but charming internet doodle style, exhausted office goblin melting into an office chair while holding a tiny coffee cup, absurd tiny-problem energy, wildly exaggerated defeated expression, messy desk chaos without readable screens, thick sketch lines, off-white paper texture, muted green accents, obvious blank space for optional caption, no words, no letters, no watermark" \
  --param aspect_ratio=4:3 \
  --param resolution=2k \
  -o funny-meme-drawing.png
对于搞笑手绘表情包，字幕为可选内容。如果仅通过绘画就能传递笑点，可直接交付图像；如果需要精确文字来强化笑话，后续再通过本地添加字幕。

4. Overlay exact text locally

4. 本地叠加精确文本

Preferred order:
  1. existing local image toolchain already used by the repo or operator
  2. simple SVG or HTML/CSS card rendered locally
  3. ImageMagick if installed
If no local renderer is available, create a simple SVG with:
  • bold uppercase title text
  • stroke or shadow for contrast
  • controlled padding and line breaks
优先顺序：
  1. 仓库或操作者已在使用的本地图像工具链
  2. 本地渲染的简单SVG或HTML/CSS卡片
  3. 已安装的ImageMagick
如果没有本地渲染工具，可创建一个简单SVG，包含：
  • 加粗大写标题文本
  • 用于增强对比度的描边或阴影
  • 可控的内边距和换行

5. QA the output

5. 输出质量检查

Check:
  • exact caption text matches the requested copy
  • line breaks read well on mobile
  • subject and caption do not compete visually
  • punchline is legible in a thumbnail
  • the output still looks intentional without knowing the prompt
If needed, use AnyCap image reading to inspect the result:
bash
anycap actions image-read \
  --file ./final-meme.png \
  --instruction "Read the visible text and describe whether the caption is easy to read at small size."
检查内容：
  • 精确字幕文本与请求内容完全匹配
  • 换行在移动端显示效果良好
  • 主体内容与字幕不会在视觉上相互干扰
  • 笑点在缩略图尺寸下仍清晰可读
  • 即使不知道原始提示词，输出内容仍看起来意图明确
如有需要，可使用AnyCap的图像识别功能检查结果：
bash
anycap actions image-read \
  --file ./final-meme.png \
  --instruction "Read the visible text and describe whether the caption is easy to read at small size."

6. Deliver

6. 交付

  • return the local file path when the human is in the same workspace
  • upload to Drive when they need a share link
  • publish a simple Page when the deliverable is a gallery or mini use-case report
  • 当用户处于同一工作区时，返回本地文件路径
  • 需要共享链接时，上传至Drive
  • 交付物为图库或小型用例报告时，发布至简单Page页面

Use-Case Article Angle

用例文章方向

This skill supports workflow-led articles better than template-library articles.
Good article angles:
  • how to make memes online with an AI agent
  • funny meme drawings with an AI agent
  • easy memes to draw with an AI agent
  • how to add text to a photo with an AI agent
  • how to make a meme video with an AI agent
  • how to create memes on Instagram without switching tools
Bad article angles:
  • obscure meme-name pages
  • "blank template" databases
  • trend-chasing pages that need constant pop-culture maintenance
本技能更适合支持工作流导向的文章，而非模板库类文章。
优质文章方向：
  • 如何借助AI Agent在线制作表情包
  • 借助AI Agent创作搞笑手绘表情包
  • 借助AI Agent轻松绘制表情包
  • 如何借助AI Agent为照片添加文本
  • 如何借助AI Agent制作短视频表情包
  • 无需切换工具，在Instagram上创建表情包
不佳文章方向：
  • 冷门表情包名称介绍页面
  • “空白模板”数据库
  • 需要持续跟进流行文化的追热点页面

Output Expectations

输出预期

A good run should usually produce:
  • 1 to 4 base visual variants
  • 1 exact-text final image
  • optional share link or published page
  • a short note explaining model choice and workflow
一次成功的执行通常应生成：
  • 1至4个基础视觉素材变体
  • 1张带有精确文本的最终图像
  • 可选的共享链接或已发布页面
  • 解释模型选择和工作流的简短说明

Guardrails

防护规则

  • Avoid copyrighted characters or branded logos unless the user provides a clear right to use them.
  • Do not promise exact text rendering from the model itself.
  • Prefer exact text overlay locally when the copy matters.
  • Avoid adult or hateful meme requests.
  • Treat meme style as a delivery format, not an excuse for sloppy output.
  • 除非用户明确提供使用权限，否则避免使用受版权保护的角色或品牌标志。
  • 不要承诺直接通过模型实现精确文本渲染。
  • 当文本内容很重要时，优先选择本地精确文本叠加。
  • 拒绝成人向或仇恨类表情包请求。
  • 将表情包风格视为一种交付格式，而非输出粗糙内容的借口。