Meeting tools already write AI summaries, but they are walls of bullet points that most of the team never opens. Comic generators exist too, but they make jokes and stories, not records. Meeting Comic sits between them: every panel is built from a real decision or action item in the transcript, with a receipt back to the exact line, in a format people read in 40 seconds and actually forward.
The data it runs on
Transcripts from the meeting tool the team already uses: Zoom cloud recordings produce a downloadable VTT transcript on paid plans; Google Meet saves transcripts to Google Docs on supported Workspace editions; Microsoft Teams lets you download a transcript as .docx or .vtt; most note-taker apps export plain text.
How it is accessed: paste or upload in the MVP. Later, an OAuth integration fetches the transcript when the meeting ends, with the host's permission.
Limits and gaps: speaker names are sometimes "Speaker 1"; crosstalk breaks lines; many free plans have no transcript at all; tone and sarcasm are easy to misread.
Fallback: paste typed notes, or upload the meeting audio and transcribe it with Whisper. Unknown speakers can be named once and are remembered.
1. The mission
Make sure everyone knows what was decided, even the people who skipped the meeting. Meeting Comic turns a transcript into a six-panel comic with the real decisions in speech bubbles and the action items in a strip at the bottom.
2. Why people can't resist it
Seeing yourself drawn. Everyone looks for their own character first, and reads the whole page to find them.
Fast to read. Six panels beat six paragraphs. Pictures carry who, what and when at a glance.
Running jokes. The same cast every week becomes a team series; the sync bug becomes a recurring villain.
3. How it works for real
Clean: parse the transcript into speaker turns with timestamps; merge short fragments.
Extract: a fast cheap model (Gemini Flash or DeepSeek via OpenRouter) labels each turn: decision, action item (owner, task, due), blocker, chatter.
Script: a frontier LLM (Claude or GPT) picks the 6 strongest moments and writes captions and bubbles, using only words that appear in labelled turns, each tied to a line ID for the receipt.
Draw: free tier uses this page's SVG character kit (zero render cost). Paid tier uses an image model (Flux or Gemini image) with a saved character sheet per teammate so faces stay consistent.
Letter and post: bubbles are typeset as live text (searchable, translatable), then posted to the team channel through its webhook.
4. Cost per use
About $0.20 per comic with painted panels: roughly $0.02 for extraction and script, plus about $0.03 per image for 6 panels. The SVG style shown here costs about $0.02 in total.
Free: 4 SVG comics a month for one team. Team ($8 a month per team): unlimited comics, painted styles, auto-posting after every meeting, the archive and custom character sheets.
5. Growth loop
Every comic posted to a channel is seen by the whole team and anyone it is forwarded to. The footer reads "Drawn by Meeting Comic", and the people in it usually want one for their other meetings: the PTA, the band, the five-a-side group.
6. Risks and guardrails
Privacy: transcripts processed through zero-retention model endpoints and deleted after 24 hours; only the comic is stored.
Consent: attendees are told a comic will be made; anyone can remove themselves, and appear as a generic figure.
Accuracy: bubbles only paraphrase labelled lines; every panel has a receipt link. Nothing is invented.
Respect: characters are chosen by each person from a kit; no caricature of bodies, accents or appearance.
7. MVP build order
Paste box plus a parser for "Name: text" and VTT.
Claude prompt that returns 6 panels and action items as JSON with line IDs.
SVG character kit and three styles (this page).
Action strip with checkboxes and a share image export.
Webhook posting to the team channel, then the archive.