Tag Category Strength (Raffle)
TagCategoryStrength is batch weight-editing for Danbooru prompts
- Adjusted tags
- Debug info
Editing prompt weights one tag at a time is fine when it's one tag. It's a drag when you realize all your artist tags are getting lost in the noise, or the meta tags are eating tokens a description could use. TagCategoryStrength lets you adjust whole families at once: give it (artist:1.4) and every artist tag in your prompt gets boosted; (meta:0.5) and the meta tags get halved. Batch weight-editing, in the format the pack already speaks.
How it works: it splits your input into individual tags, looks each one up in the pack's bundled categorized_tags.txt (tens of thousands of tags mapped into categories like poses, clothes_and_accessories, artist, meta, and the rest), and whenever a tag's category matches one of your adjustments, it rewrites the tag with standard ComfyUI weighted syntax - (poses:1.2), (artist:1.4), and so on.
The inputs:
input_tags- comma-separated tags. It's force-input, so you can wire it straight from another node - in practice, theRaffled outputfrom this same pack.category_adjustments-(category:strength)pairs, comma-separated, e.g.(artist:1.4), (meta:0.5), (poses:1.2).preserve_existing_weights(default true) - if a tag already carries a weight like(tag:1.2), keep it instead of re-weighting. Turn it off if you want the category strength to multiply over existing weights.
Outputs: Adjusted tags, which wires into a CLIP Text Encode, and Debug info, which tells you what category each tag landed in. That second output is more useful than it looks - it's how you'll discover a tag you thought was a pose is actually filed under clothes_and_accessories.
Two gotchas:
- Category names are validated hard. Mistype one in
category_adjustmentsand the node raises an error listing every valid category. The Raffle node's Debug output shows the full list if you need it. - Weighted syntax is a CLIP-encoder thing.
(tag:1.4)means something on Illustrious, NoobAI and Pony, and nothing on an LLM-encoded model, where weights are just literal punctuation passed through to the encoder. This node is for the SDXL anime family - which is, to be fair, exactly the family the whole pack targets.
It's a small node with a small job, and it's the natural finishing move after a Raffle pick: roll a taglist, then push the parts you actually care about without hand-editing forty tags. Install is the pack, same as the others:
cd ComfyUI/custom_nodes
git clone https://github.com/rainlizard/ComfyUI-Raffle
...or "Raffle" in ComfyUI Manager. No extra dependencies.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| input_tags | STRING | Input tags to adjust (comma-separated) | |
| category_adjustments | STRING | Category adjustments in format: (category_name:strength), e.g., (artist:1.4), (meta:0.5), (poses:1.2) | |
| preserve_existing_weightsopt | BOOLEAN | true | If enabled, tags that already have weights like (tag:1.2) will keep their existing weights instead of being adjusted |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| Adjusted tags | STRING | Tags with category-based strength adjustments applied |
| Debug info | STRING | Information about which tags were adjusted and their categories |