Markdown转图片 / Markdown To Image

Your LLM node just handed you a wall of Markdown and you need it as an image - a card for Xiaohongshu, an A4 printout, a 1200×630 link cover. That's this node's whole job: feed it Markdown, get a rendered IMAGE out, with typography that actually looks like a reader, not a text dump pasted onto a canvas.

It comes from RUI-Nodes (Rui-Node🐶), a Chinese-authored utility pack that mostly lives on sprite tooling and matting. This text-side node is one of its friendlier corners, and it's pure PIL - no model weights, no GPU, no hidden downloads.

What it actually renders

The visual spec is deliberately GitHub / Typora-flavored. Headings scale 2.0/1.5/1.25/1.0/0.875/0.85× the body size, H1 and H2 get the divider rule, blockquotes get the left bar, code blocks get rounded corners plus a language tag, and tables come with the header fill, zebra rows and column alignment you expect. Task-list checkboxes and colored emoji work too - emoji render via your system's color fonts, so no mojibake squares.

Everything sits on pure PIL, which is the clever part: no browser, no Playwright, no HTML-to-image dependency chain to babysit. The size fields (body_size, h1_size…h6_size, letter_spacing, line_spacing, max_chars_per_line) default to auto, and the node binary-searches the font size to "just fill" your canvas instead of guessing. It'll shrink the font to fit rather than crop your content.

The inputs that matter

• markdown - paste straight in. Multiline, so a whole report fits.
• size_preset - nine quick picks: 1080×1440 (3:4), 1080×1920 (9:16), 1080×1080, 1920×1080, 1280×720, 1200×630 link cover, A4 portrait, and more. Pick custom and width/height (64–8192px) only when you need something exact.
• font - drops from the Ruinode/font directory. Drop a ttf/otf/ttc in there, refresh the page, it appears. A same-family bold file (like msyhbd) pairs up automatically for bold text; without one it fakes bold with a stroke.
• theme - light / dark / sepia. Dark reads well for code-heavy cards.

Output: one image (IMAGE), sized to whatever you picked.

The trap everyone hits: WAS Text Multiline

The author is explicit in the tooltip: don't feed this node from WAS Node Suite's "Text Multiline". That node treats #-prefixed lines as comments and deletes them - your headings vanish silently - and it also does dynamic-prompt substitution. If you need a text carrier upstream, use this pack's 多行文本框(原样输出) / Text Box (Raw) node instead, which passes text through verbatim.

Installing

It's part of RUI-Nodes, so the whole pack installs at once:

ComfyUI Manager → search "RUI-Nodes" → install, restart
# or manually:
cd ComfyUI/custom_nodes
git clone https://github.com/rui40000/RUI-Nodes
cd RUI-Nodes && pip install -r requirements.txt

Then restart ComfyUI. The requirements file pulls torch, numpy, Pillow and requests, plus diffusers/transformers for the pack's matting nodes - a stock ComfyUI environment usually already has most of it. This node itself adds no dependency beyond PIL.

If a heading shows up missing, it's almost never the renderer - check what's upstream of the markdown input. And if the output crops text at the bottom, shrink the body size or raise the canvas instead of chasing a phantom setting.