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

The Style & Prompt Encoder family is the soul of Z-Image Power Nodes - Martin Rizzo hand-tuned a hundred-plus styles for Z-Image's natural-language prompting - and this experimental member is the one where you stop picking his styles and start writing your own, without leaving the node to edit a text file.

Custom Style + Prompt Encoder ^G2.1 is the custom-definition variant of StylePromptEncoderX21. Instead of drawing from the pack's built-in style library, the styles live in a multiline string you type (or wire in) yourself. Same encoding mechanics, same optional color palette, but the style list is yours. If you've ever spent an afternoon fighting a .sty file to get one look right, this is the workflow-friendly version of that.

How it works

The style definitions use a simple block format. Each one starts with >>> followed by the style's name, then the description on the following lines, with {$@} marking where your base prompt gets inserted:

>>> Loose Watercolor
Render the scene as a loose watercolor painting, soft edges,
paper texture visible, colors bleeding gently at the borders.
{$@}

You write the style once as the custom_styles input, then flip between your styles through the custom_style dropdown. The node finds your definition, applies the template to your prompt (substituting {$@}), optionally layers in a color palette, and runs the result through your CLIP - here, the Qwen3-4B text encoder that Z-Image uses - to produce conditioning.

The inputs that matter

• custom_styles (STRING, multiline) - your style library in >>> format. This is the whole point of the node.
• custom_style (selector) - which of your custom styles to apply.
• prompt (STRING) - your base text.
• palette (optional) - feed a PaletteSelector output here to add a color scheme description to the prompt.

The outputs are unusually chatty and it's great: CONDITIONING (wire into your sampler), PROMPT (the final text after style+palette), plus style_name, palette_name, and original_prompt - handy for debugging exactly what got injected, or for logging your generation parameters.

Install

Same pack, same drill:

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

Restart ComfyUI. (Or ComfyUI Manager → search "Z-Image Power Nodes".) No extra pip deps. You'll need the Z-Image model triplet installed - diffusion model, Qwen3-4B text encoder, and VAE - since this node is only the front end of encoding.

Worth the caveats?

Two things to know. First, it's experimental (⚗️🔬) - the author flags that experimental nodes can change or disappear without notice, so save your style definitions somewhere outside the workflow. Second, the G2.1 styles and palettes are a newer format than the stable G2 line; the built-in style list in the stable Style & Prompt Encoder is v1.0.0 while this experimental branch reads v2.0.0 definitions. If you're starting fresh and don't yet have your own style library, start with the stable encoder and come here once you know exactly what look you're chasing.