anycap-social-meme-workflows（Anycap 社交模因工作流程）

AnyCap Social Meme Workflows

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.

Read First

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.

Best Fit

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

Core Rule

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

Workflow

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

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

2. Choose the model

Default mapping:

• Seedream 5 for stronger first-pass visuals
• Nano Banana Pro when editing an existing image or screenshot
• Nano Banana 2 when you need many variants fast, especially for funny meme drawings

Always inspect the model list or schema if the workflow is unclear:

anycap image models
anycap image models nano-banana-2 schema

3. Generate or edit the base visual

Text-to-image example:

anycap image generate \
  --model seedream-5 \
  --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:

anycap image generate \
  --model nano-banana-pro \
  --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 with Nano Banana 2:

anycap image generate \
  --model nano-banana-2 \
  --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.

4. Overlay exact text locally

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

5. QA the output

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:

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

• 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

Use-Case Article Angle

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

Output Expectations

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

Guardrails

• 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.