Erics Word Cloud Generator (Advanced Image/Mask Support)

Let's be honest: a word cloud node is not why you installed ComfyUI. You installed it to make images, and this pack is mostly a metadata system. But Erics Word Cloud Generator is the pack's most fun outlier - it takes a blob of text and produces a proper word-cloud image, as an IMAGE tensor, right in your graph. It's built on the classic Python wordcloud library, and it turns out ComfyUI is a surprisingly good home for it, because you can pipe the result straight into a sampler or an upscaler, or export it as real vector SVG for printing.

The required inputs are the sane ones: text (multiline - paste a speech, a paper abstract, or a dump of your favorite Civitai tags), width/height, max_words, min_font_size/max_font_size, and background_mode (transparent or solid_color) plus background_color. Hit Queue and you've got a cloud. The power lives in the optional section, and this node is deep: seven color modes (single color, random, cultural sets like red_white, emotional sets like energetic, harmony sets, custom ranges, matplotlib colormaps like viridis), four font modes, normalize_plurals (combines "cat"/"cats"), collocations for two-word phrases, custom stopwords, and special_words for boosting specific terms via JSON with a size_multiplier.

The five outputs are where it becomes genuinely useful: wordcloud (IMAGE, the cloud itself), font_preview and color_preview (IMAGE, so you can eyeball fonts and palettes before committing), svg_export (STRING - real vector SVG markup), and used_words (STRING, so you can see what actually made the cut). The intended loop: tune in the font_preview, then save the wordcloud with the pack's Save Image with Metadata nodes, or feed svg_export into the pack's Save True SVG Image node for a lossless vector file. There's even a mask_image input (MASK) with mask_placement to force words inside a shape, plus input_image as a background to overlay on.

Mechanically it's the wordcloud package doing the layout; this pack just wires it into ComfyUI types and wraps it in a genuinely enormous set of options. Dependency note: wordcloud is in the pack's requirements.txt as required, not optional, and the node's font list is pulled from your system fonts - which is why you'll see DejaVu and C059 in the dropdown rather than a curated set. If the dropdown ever shows an error string instead of inputs, the wordcloud package didn't install; pip install wordcloud fixes it.

Practical takes: random_seed is your friend for reproducible clouds, relative_scaling (default 0.8) balances frequency-vs-rank sizing, and the special_words tooltip is honest that the library can't do positioning, so don't fight it. This is a one-author pack with zero community chatter, so treat the node as self-serve - but for a tool that sounds like a toy, the output quality is genuinely decent.

Install the pack and restart:

cd ComfyUI/custom_nodes
git clone https://github.com/EricRollei/AAA_Metadata_System
cd AAA_Metadata_System
pip install -r requirements.txt

Feed it your text, pick a colormap, and try not to spend an hour on it. You will anyway.