🪴 Brain

OpenClaw Meme Maker

3 min read

OpenClaw Meme Maker

🗒️ The meme-maker skill in openclaw/openclaw is a small MIT skill (3 files) that turns a joke topic into a captioned meme. It ships a curated registry of about 20 templates with named fields, hand-placed text boxes and a “use” line each, and it does not ship copyrighted template images: in local mode it downloads and caches them, writes SVG and rasterises to PNG with Playwright. An Imgflip mode is also available. It is the closest thing to a ready design reference I found for an agent-picks-template skill, and the cheapest base to fork in the Meme Generation for Coding Agents plan.

Description

  • Workflow. node scripts/meme.mjs suggest "<topic>" --json (a keyword ranker) or read references/templates.json directly; the agent chooses the template and writes one caption per named field (drake: “rejected option”, “preferred option”); then render drake --text "…" --text "…" --out ./meme.png. Imgflip mode adds --service imgflip and prints a hosted URL. One to two tool calls.
  • Tested by the research agent: an English topic (“deploy on friday broke production”) ranked this-is-fine well; Polish diacritics survived in the drake SVG and a four-panel expanding-brain PNG, upper and lower case, no forced uppercase. SVG took about 0.3 s and PNG about 5.7 s through playwright-core with the installed, signed Chrome.
  • Activity. The repo is huge (about 392k stars) and the skill has had four commits (2026-05-17 to 2026-10-08); skills.sh shows 322 installs. [3rd-party]

Download or use

Node 18 or newer. Copy skills/meme-maker/ into ~/.claude/skills/meme-maker/ (the skill uses OpenClaw’s {baseDir} placeholder, so substitute the folder path or let the agent do it). For PNG add npm i playwright-core next to the script and have Chrome or Chromium installed; never run playwright install, its downloaded ffmpeg was blocked by Smart App Control. Edge alone gives SVG only.

Reasoning for

It fits my constraints better than anything hosted: local rendering, no account, no watermark, a registry that tells the agent what each format is for. Known limits, all fixable in a fork:

  • Captions wrap to at most 5 lines and overflow is silently dropped; extra --text values reuse the last box and overlap.
  • suggest strips non-ASCII, so Polish topics rank by accident (the agent picks the template anyway, so this matters little).
  • The SVG depends on the viewer’s Impact font; there is no custom-background option.
  • Template images are hot-linked from i.imgflip.com and can disappear; Imgflip mode takes only a username and password, and credentials passed as flags leak into transcripts.
  • Historical incident: an earlier error message told agents to install sharp, an agent did so inside a pnpm-managed OpenClaw tree and pruned 379 node_modules entries (issue #109405). Strip any “install this” advice from a fork. See AI Agent Security.

My plan is to seed my own catalog from its registry and swap the renderer for Satori so PNG no longer needs a browser.

Alternatives considered

  • Own skill on Satori + resvg-wasm (the design in Meme Generation for Coding Agents): same idea without the browser dependency and with a tofu detector for Polish glyphs.
  • Meme Factory: a skill on Memegen, simpler, but 8 of its 21 template IDs are dead and it hides the watermark.
  • Hosted MCPs (Memesio, Createa): less control, watermarks, anonymous operators.

Resources

Template: tool

Graph View

Backlinks