Excalidraw Writing Machine: Generate articles from your Visual Zettelkasten cards in Obsidian

A workflow built around Visual Zettelkasten cards in Obsidian can turn a visual article layout into a clean, publication-ready PDF draft—complete with section structure, embedded images, and source citations. The core idea is to treat each visual card as a Zettelkasten note (front: drawing/summary; back: markdown text, sources, and supporting notes), then use an “Excalidraw Writing Machine” script to read the card sequence and automatically “stitch” the content into a structured document.

The promise behind Zettelkasten is familiar: once a system has accumulated notes, new writing shouldn’t start from scratch. Instead, writers should be able to assemble existing notes into a first draft and then refine. This approach adds a visual layer to that promise. The framework described breaks personal knowledge management into dimensions: linear text (read top-left to bottom-right), separate documents, links between notes, and—crucially—illustrations reused across notes. Visual processing is treated as parallel rather than linear, with icons and image fragments acting as connective tissue between ideas.

Visual Zettelkasten operationalizes that by giving each note two sides. The front side is the visual representation (drawings in Excalidraw), while the back side holds markdown content such as the idea summary, related notes, and sources. When organizing ideas, the system doesn’t just shuffle bullet points; it organizes at the card level. That card-level structure is what the writing machine script translates into an article draft.

Technically, the script walks through an Excalidraw “arrow sequence.” It captures the main title, then paragraphs until it hits a branch point. At each “fork in the road,” the first text element becomes a section heading, followed by the paragraphs that flow linearly from that point. When a Zettelkasten node appears in the layout, the script pulls the card’s back-side markdown (including the source reference) and places it under the corresponding image in the generated output. The result is a document with headings, paragraphs, images, and citations formatted into a coherent draft.

From there, the workflow offers two automation levels. In the lighter mode, the generated markdown can be copied into ChatGPT with a prompt to rewrite and expand it into a full article. In the heavier mode, AI4 Templator and Templator are used to run ChatGPT queries automatically inside Obsidian. The script passes generated text via a window variable (window.wm), supports debugging by storing responses back onto the window object, and uses a placeholder in the template so the original extracted text can be injected into the AI prompt.

Installation is handled through Obsidian’s “install/update scripts” flow for Excalidraw scripts, and the script’s settings allow customization such as renaming the summary section, selecting where sources live, and choosing whether to embed images. Because the solution is a prototype, it’s designed to be edited: the arrow interpretation depends on drawing order, and misordered connections can reorder sections in the output. The practical takeaway is that visual organization becomes a structured drafting engine—turning a visual knowledge map into a readable document that still requires human polishing before publication.