2🐕Quality category

"masterpiece, best quality, ultra detailed" on one side, "worst quality, blurry, watermark" on the other - nearly every SD 1.5/SDXL-era prompt template has some version of this boilerplate, and typing it fresh every time is pure friction. EG_TSCDS_ZL ("Quality category") packages both the positive quality tags and the negative-prompt boilerplate into one dropdown-based node, so you're picking from a curated list instead of retyping from memory.

How it works

Four categories in one node: Image type (highres, absurdres, official art, scan, 4koma…, 91 choices), Renderer (Unreal Engine, Octane, Corona Render, V-Ray, UE5…, for renders that want to look CG-authored rather than photographed or drawn), Positive prompt word (best quality, masterpiece, ultra high res, intricate detail, 8K…), and Negative prompt word (worst quality, lowres, blurry, easynegative, watermark, artist name…). All four combine into one output string, so in practice you'll use this node twice per workflow - once with only the positive-leaning fields filled, routed to your positive CLIP Text Encode, and once with only the negative fields filled, routed to your negative encode. It doesn't split positive from negative for you; that's on your wiring.

The inputs and outputs that matter

Each of the four categories carries its own weight slider (0.1–2.0, default 1) and defaults to "None." The tested useful range for tag emphasis sits around 0.5–1.5 - past that you're inviting artifacts rather than a cleaner result, which matters especially for the quality-booster tags since it's tempting to stack several at high weight thinking "more quality tags = better image." random (default no) plus seed roll a random selection across all four categories instead of your manual picks. Output is a single prompt (STRING).

How to install it

Search Comfyui-ergouzi-Nodes in ComfyUI Manager, or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/11dogzi/Comfyui-ergouzi-Nodes.git

Restart. As with the rest of this pack, the author's README marks this English repo as frozen, pointing readers to a more actively maintained Chinese-named sibling for new features.

Common issues & troubleshooting

Quality-tag boilerplate doesn't do much on modern checkpoints. "masterpiece, best quality" was a workaround for how SD 1.5 and early SDXL models were trained and tagged - it's much less effective, sometimes a no-op, on newer or differently-trained checkpoints, and on LLM-encoded models (Z-Image, Flux 2, Anima, Krea 2) it doesn't function as intended at all, since those models weren't trained on stacked booru-style quality tags the same way.

Weight syntax breaks entirely on LLM-encoded models. Same caveat as the pack's other prompt-word nodes: (masterpiece:1.3) reads as emphasis only on CLIP-based checkpoints. On an LLM-encoded model the parentheses and colon come through as literal characters in your prompt - actively worse than typing the tag plain.

You accidentally combined positive and negative tags in one output. Because this node doesn't separate the two categories at the output level, filling in both Positive prompt word and Negative prompt word on the same instance produces one string containing both - which will confuse whichever single encode you wire it to. Use two separate instances of this node, one per polarity, if you need both.