Rating, Score, Source for Pony

If you've generated with Pony Diffusion, you've run into the score tag ritual: to get good output you don't just add score_9, you add the whole cumulative chain - score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up. That's not a stylistic choice, it's a training artifact. AstraliteHeart's team labeled training images with quality scores intending you to be able to say "everything score_8 and up," but a training quirk meant the model learned the literal string of tags together rather than each one independently - so you need the full chain, not just the top tag, or quality drops. This node exists so you never have to remember that, or type it by hand.

What it builds

Three dropdowns, one output. score covers the quality end (from the description's own example: things like -, Everything, Average, Good, Only the best), and picking a tier is what constructs the correct cumulative score_N_up chain behind the scenes. rating picks Pony's content-rating tags (Safe, Questionable, Explicit), and source picks its style/source tags (Anime, Furry, Cartoon, Pony) - these steer which visual tradition the model leans into, since Pony's dataset spans booru anime art, furry art, and western cartoon styles side by side. The - option on each dropdown is the "skip this category" choice, so you can build a prefix using only the axes you actually care about. All three selections get combined into one combined_string output, comma-separated and ready to drop straight into your positive prompt or a text-concat node ahead of your actual subject description.

Why this beats typing it yourself

Beyond just saving keystrokes, this sidesteps the two easiest ways to get Pony prompting wrong: forgetting the cumulative score chain (using just score_9 alone genuinely under-performs), and mixing source tags that fight each other - stacking source_anime in the positive with source_furry left out of the negative, for instance, when the community convention is closer to putting your target source in the positive and the others in the negative to steer style harder.

The one thing worth knowing going in: these tags are Pony-specific. They do nothing useful on Illustrious, NoobAI, SDXL base, or Flux, even though those models share plenty of Danbooru-lineage habits with Pony. If you're prompting a different base model, this node's output is dead weight in your prompt, not a helpful convention.

How to install it

ComfyUI Manager: search DemonAlone-nodes-ComfyUI, install, restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/DemonAlone/DemonAlone-nodes-ComfyUI

then restart. No models, no extra dependencies - it's a string builder, live as soon as the pack loads.

Common issues & troubleshooting

I'm getting muddy or generic results even with the right tags. Score and rating tags steer quality and content, but they don't fix a bad base prompt. Pony's booru-tag training means it wants dense, specific tags describing the actual subject - natural-language sentences work much worse on Pony than on SDXL base, regardless of what this node contributes up front.

These tags aren't doing anything. Confirm you're actually on a Pony-based checkpoint (V6 XL or a Pony merge), not Illustrious or another SDXL finetune that superficially resembles it - the score/rating/source vocabulary is specific to Pony's training and won't transfer.

Should I always max out score? Only the best biases toward Pony's highest-quality training tier, which is the generally recommended default - dropping to a lower tier on purpose is mostly useful if you're deliberately exploring a rougher or more stylized look, not something most people want by default.