◒ Pony Tokens

This node needs an honest caveat up front, and it's worth reading before you wire it in: the score_9, score_8_up, score_7_up... quality vocabulary it's built around is Pony Diffusion's, and it does essentially nothing on a straight Illustrious, NoobAI, WAI, or Hassaku checkpoint. That's not a limitation of this node - it's how Pony's training worked. A training-time bug fused the whole score string into one learned quality signal specific to Pony's own dataset, and none of it transfers to a model that never saw those tokens during training.

So who is this node actually for? Checkpoints that are genuine Pony/Illustrious merges - and these exist and are labeled as such on model-sharing sites - carrying real Pony training underneath an Illustrious-flavored finish. It's also useful if you're deliberately porting or testing a prompt across model families and want the Pony-side vocabulary available without retyping it from memory.

What each field maps to

Quality translates Pony's numeric score tags into readable buckets - the tooltip gives GOOD ≈ score_7_up as a reference point, so you don't need to remember the exact score cutoffs; EVERYTHING (the default) is the least restrictive setting. Source and Rating map directly to Pony's source_* and rating_* tags, which steer style family (anime/pony/furry/cartoon) and content rating respectively - these are the parts of Pony's tag convention with the most standalone reason to exist, since rating control is genuinely useful vocabulary on any checkpoint that actually learned it.

Invert Source (Neg) and Invert Rating (Neg) push the chosen tag into the negative prompt instead of the positive - the standard Pony-side move is pairing a positive source tag with the unwanted sources negated, e.g. source_anime positive with source_furry, source_cartoon negative, to keep the model locked into one aesthetic family rather than drifting between them.

Pony Quality Boost (Beta) and Pony Negative Boost (Beta) append curated tag lists on top - both explicitly marked Beta in their own tooltips, so expect them to shift between pack releases.

The inputs and outputs that matter

• Quality, Source, Rating - the core three, all enums, all with a neutral -/EVERYTHING option so you can leave any of them off.
• Invert Source (Neg) / Invert Rating (Neg) - booleans, off by default.
• prefix / suffix (optional) - splice extra text around the generated tokens.

Outputs: PROMPT and NEGATIVE, both strings - merge these into your existing positive and negative prompt chains rather than treating them as a full replacement.

Installing it

ComfyUI Manager: search "EasyIllustrious," install, restart. Manually:

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

No downloads beyond the pack itself.

Common issues

These tokens seem to do nothing at all. Check your checkpoint first, before assuming the node is broken. On a pure Illustrious-lineage model with no Pony training, this is expected - the vocabulary simply isn't in that model's learned space. Confirm you're actually on a documented Pony/Illustrious merge before troubleshooting further.

Stacking this with the pack's Danbooru-side quality tags (masterpiece, best quality) and getting inconsistent results. That's two separate quality vocabularies from two different model families layered in one prompt - on most checkpoints only one of them is doing real work. If you're not specifically testing cross-compatibility, pick the vocabulary that matches what your checkpoint actually trained on and drop the other.