Meme Card Maker

Meme-format graphic generator. Input: top text + bottom text + optional centerpiece (template or user photo). Output: classic meme-aesthetic image with Impact-style typography.

Distinct from quote-card-maker:

• Two text blocks (top + bottom), not one quote + attribution
• Impact-font bold-with-stroke typography (the universal meme look)
• Often references a known meme template (drake, distracted-boyfriend, expanding-brain)
• Aesthetic: meme template, NOT editorial — uglier is often funnier

Distinct from flyer-maker / cover-maker:

• No event details, no creator metadata
• Different aesthetic vocabulary (Impact font, stroke outline, low-fidelity feel)
• Often references specific cultural meme formats

This skill does NOT:

• Generate animated GIF memes — use gif-maker separately
• Watermark / credit the original meme template creator — that's your responsibility
• Verify cultural appropriateness — your judgment
• Translate meme captions — pass in target language
• Place text at PIXEL-perfect template positions — model approximates

ROLE

Read top text + bottom text + optional template hint + optional base photo → pick visual-integration-strong model → assemble prompt cueing Impact-style typography with stroke, meme aesthetic → batch execute → save PNGs.

PIPELINE

1. Resolve text:

• --top "" — top caption (≤8 words, ALL CAPS rendered)
• --bottom "" — bottom caption (≤8 words, ALL CAPS rendered)
• At least one of top / bottom required

1. Resolve template (optional):

• --template drake|distracted-boyfriend|expanding-brain|two-buttons|change-my-mind|custom
• drake: 2-panel "rejects / approves"
• distracted-boyfriend: 3-character composition
• expanding-brain: 4-panel ascending
• two-buttons: sweaty decision
• change-my-mind: seated man with sign
• custom: just the captions on whatever centerpiece the model generates
• Default: custom

1. Resolve base photo (optional):

• --base-photo — use user image as centerpiece
• Implies --template custom (template hints are ignored when base photo is provided)

1. Pick model:

• With --base-photo: nano-banana-pro (preserves visual, adds caption layer)
• Template-based: gpt-image-2 (best at illustration + text rendering)
• Override: --model

1. Compose ONE LLM call — load [common/visual-prompt-library/system-prompt.md](../../common/visual-prompt-library/system-prompt.md) (shared SYSTEMPROMPT) and buildUserMessage(opts) with Mode=meme-card, N=, top + bottom captions + template + optional base photo. Spawn ONE Agent. Receives JSON {"slides":[{"number":1,"prompt":"..."}]} — 1–3 sentence prompts with template-specific composition + Impact-style typography cues + meme aesthetic ("white text with thick black stroke, all caps, low-fidelity, slight JPEG artifacts"). Captions in double quotes. See references/templates.md for template composition. Library styles via [common/visual-prompt-library/styles/_index.md](../../common/visual-prompt-library/styles/index.md).

1. Estimate cost + confirm — inherits SKILLS_CAROUSEL_BUDGET=1.50.

1. Batch execute — common.runners.batch.run_batch().

1. Output:

`` ./generated/meme// meme-v1.png meme-v2.png manifest.json prompts.md ``

MODES

Required

• meme-card-maker --top "" (top) OR --bottom "" (bottom; at least one required)

Optional content

• --template drake|distracted-boyfriend|expanding-brain|two-buttons|change-my-mind|custom
• --base-photo — user image as centerpiece
• --context "" — situational context to help the model interpret captions
• --lang en|ru — language hint (default: en)

Visual

• --variants N — variants (default 3)
• --model auto| — image provider
• --style-mod "" — append tweak (e.g., "extra pixelated, deepfried meme aesthetic")
• --aspect square|portrait|landscape — composition aspect (default square 1080×1080)

Execution

• --execute — actually generate
• --output — custom output
• --parallelism N — concurrent calls (default 2)
• --yes — skip cost confirmation
• --resume — retry failed
• --prompts-only — dry run

REFERENCES (load on demand)

| File | When to load | |---|---| | [references/templates.md](references/templates.md) | Step 2 — meme template anatomy, composition cues per template | | [references/typography.md](references/typography.md) | Step 5 — Impact font conventions, stroke outline rules, when to deviate | | [references/troubleshoot.md](references/troubleshoot.md) | When text doesn't read, template doesn't recognize, etc. |

EXAMPLES

See [examples/before-after.md](examples/before-after.md) — 3 calibration runs: drake template developer joke, custom photo meme, expanding-brain progression.

CONSTRAINTS

• Captions ≤8 words each. Past 8, text overflows; Impact-style bold needs space.

• ALL CAPS rendering. All meme captions auto-uppercase. If you pass lowercase, it's rendered uppercase. Exception: --lang ru keeps mixed case for Cyrillic readability (Cyrillic ALL CAPS is harder to read).

• Default model is gpt-image-2. Best at illustration + integrated text. nano-banana-pro when base photo is provided.

• Template recognition is approximate. "Drake" template is approximated by the model — won't be pixel-identical to the original. For exact template: use a meme generator tool (Imgflip, Memegenerator).

• Cultural sensitivity is YOUR responsibility. The skill doesn't filter. Don't ship punchy memes about sensitive topics without consideration.

• Generated memes can resemble copyrighted templates. Original meme creators rarely enforce, but for commercial use: check the template's original source.

• Cost confirm ONCE per batch. Sum across variants.

• Aspect default is square. Most platforms display memes 1:1 well. --aspect portrait for IG portrait posts; --aspect landscape for Twitter inline.

• Never print API keys. Mask in errors.

INVOCATION HINTS

When the user says any of:

• "make a meme", "meme for X", "meme template Y"
• "drake meme", "expanding brain meme", "distracted boyfriend"
• "сделай мем про X", "мем формат Y"
• "сгенерируй мем"

If the user names a template explicitly, set --template . If unclear, default --template custom (lets the model interpret).

If the user pastes a photo path: --base-photo + --template custom.

Defaults: --variants 3 --aspect square --template custom --model gpt-image-2. Without --execute, returns prompts.

This skill is distinct from:

• quote-card-maker — text-dominant editorial cards (single quote + attribution)
• image-prompt — free-form image gen (no meme conventions)
• gif-maker — animated content (memes here are static)
• flyer-maker / cover-maker — structured professional formats

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Mikefluff
• Source: Mikefluff/skills
• License: MIT
• Homepage: https://www.npmjs.com/package/@mikefluff/skills

Install and usage instructions live in the source repository linked above.