❌[DEPRECATED] Style + Prompt Encoder ^X2.1

Style & Prompt Encoder ^G2.1 is what happens when the pack's signature style encoder gets a v2.1 upgrade: same core idea - pick a style, get your prompt rewritten in its language and encoded into conditioning - but now with a color palette baked into the pipeline, a fresh style definition set, and a lot more visibility into what actually got encoded. It's the experimental branch of the style family, marked ⚗️🔬, and it sits next to the stable StylePromptEncoder2 rather than replacing it.

If you're choosing between the two, the short version: the stable node has the mature 100+ style library and is what you should default to. This one is where the author is testing the palette idea and a newer style format - great to experiment with, but its style set is thinner and the API is explicitly subject to change.

How it works

Same mechanics as the stable encoder: your text is wrapped in the selected style's template and encoded through your CLIP. The additions are the palette input (a ZIPN_PALETTE_SELECTOR - grab one from the PaletteSelector node) which folds a color scheme description into the prompt, and a customization input for defining your own >>> styles. The output side is where it really diverges from the stable node: instead of just conditioning plus the styled string, you get five outputs.

The inputs that matter

• clip (CLIP) - your text encoder, i.e. Qwen3-4B.
• style (selector) - from the G2.1 style set.
• palette (selector) - optional-ish color scheme from PaletteSelector.
• text (STRING, multiline) - base prompt.
• customization (optional STRING) - custom style definitions.

The five outputs

• CONDITIONING - wire into your sampler.
• PROMPT - the final text after style + palette, so you can see exactly what's being encoded.
• style_name and palette_name - which style and palette got applied, handy for logging or metadata.
• original_prompt - your untouched input, so you can always recover it downstream.

That output set is genuinely useful for debugging (and for the pack's CivitAI metadata export), and it's the best reason to try this node.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/martin-rizzo/ComfyUI-ZImagePowerNodes

Restart ComfyUI (or ComfyUI Manager, search "Z-Image Power Nodes"). No pip dependencies; needs the Z-Image model triplet to actually sample.

Caveats

It's experimental - the author's boilerplate applies: parameters, behavior, or availability may change or vanish without notice, so don't build a mission-critical workflow on it. The G2.1 style library is also younger than the stable one; if the exact look you want exists in the stable set but not here, you may have to carry your style definitions in customization. And the palette feature only lives on this experimental line, so if you don't care about color palettes or the extra outputs, the stable encoder is the safer, more complete choice.