Tag Frequency Weighter for ComfyUI

A ComfyUI custom node that weights tags based on their frequency/rarity in danbooru

Features

• Automatically weights tags based on popularity/rarity
• Caches calculations for better performance
• Preserves existing weights in the input prompt
• Configurable weight ranges and pivot points
• Filter tags by count thresholds

How It Works

1. Reads tag frequencies from danbooru_tags_post_count.csv
2. Calculates averages and caches them in tag_stats_cache.json (recomputable via toggle)
3. For each tag in the prompt:
   • Maps rarity/popularity around a pivot to your chosen [min_value, max_value] range
   • Popular tags → weighted toward min_value
   • Rare tags → weighted toward max_value
   • Tags that round to 1.00 remain unchanged

Inputs

Parameter	Type	Description
input_prompt	string	Input tags (comma-separated, preserves existing weights)
min_value	float	Lower bound for output weights
max_value	float	Upper bound for output weights
calculate_average	bool	Recompute CSV stats and update cache
preserve_underscores_in_output	bool	Keep underscores in output tags
pivot	string	log10(avg) or x_avg - calculation method
ignore_below_tag_count	int	Leave tags below this count unchanged (0=off)
ignore_above_tag_count	int	Leave tags above this count unchanged (0=off)
debug	bool	Print per-tag counts during processing

CSV Source

The dataset was created using script by bem13.

To rebuild the dataset:

1. Generate danbooru_tags_post_count.csv using the script
2. Place it next to nodes.py in the node directory

Installation

1. Drop this folder into ComfyUI's custom_nodes/ directory
2. Restart ComfyUI

License

MIT