Dataset Wordcloud

Dataset Wordcloud turns a folder of prompt text files into a wordcloud image - the classic "what does my dataset actually keep saying?" visualization. It's a diagnostic tool more than a creative one, and that's its strength. When you've collected hundreds of prompts for a LoRA or a style experiment and you want to know whether you've accidentally over-indexed on one concept, or whether your captions are dominated by a phrase you don't care about, a wordcloud makes the answer obvious in one glance.

It's genuinely useful in the training-adjacent workflow: dump your caption corpus in, see the dominant tokens, catch the redundancy. The pack's author is big on dataset tooling (the same folder has LoadTextDirectory and a bunch of prompt-engineering nodes), and this is the "stop squinting at the CSV" node.

How it works

Under the hood it's the wordcloud library doing its standard frequency-based layout - more frequent terms get bigger - rendered through matplotlib and handed back as an IMAGE tensor. Frequency is the whole algorithm; there's no sentiment analysis, no topic modeling. What you're seeing is pure term frequency, which is exactly what you want for spotting over-represented concepts.

The input is a dropdown (text_file) populated from the pack's prompts/ directory - there are 200+ .txt files shipped with DJZ-Nodes, so you get a wall of options out of the box. To analyze your own corpus, drop your .txt into that folder and restart.

The inputs that matter

• text_file - the dropdown of .txt files from DJZ-Nodes/prompts/. This is where you point it at your data.
• width / height - canvas size, 64–4096. Defaults 800×400; bump for a wallpaper.
• color_palette - five options: kandinsky, warm, cool, monochrome, vibrant. Pure cosmetics.
• background_color - a plain STRING, so you can type black or #111111.

Output is a single IMAGE (batch of one) at your requested size.

Installing it

Dataset Wordcloud ships in DJZ-Nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/MushroomFleet/DJZ-Nodes
cd DJZ-Nodes
pip install -r requirements.txt

Restart ComfyUI, or install via Manager ("DJZ-Nodes"). This node needs wordcloud and matplotlib, both in the pack's requirements. Nothing heavy, no models.

Common issues

The dropdown only shows files that existed at startup - add a new .txt and it won't appear until you restart, same story as the borders node in this pack. Second, the node writes a temp_wordcloud.png into the current working directory as part of rendering; it cleans up after itself, but if ComfyUI crashes mid-run you'll find a stray file. Third, remember it's counting raw tokens, so common filler words (the, with, high quality) will dominate unless your prompts are already clean - which is honestly a useful signal about your captions in itself.