⚡| Style String Injector

Style String Injector is the style system for people who already have a text-encoding setup they like. The style encoders in this pack bundle two jobs - rewriting your prompt and encoding it - but the Injector only does the first. You give it a style and a string, it returns the styled string, and you decide what happens next: your own CLIP Text Encode, a different pack's encoder, a prompt combination node, anything that eats text. It's the most composable member of the style family, and it's the current version (the pack's original StyleStringInjector is the deprecated predecessor).

The practical payoff: style becomes just another text transform in your graph. You can style a prompt and still tweak it, concatenate it, preview it, or feed it to conditioning for a model that isn't even Z-Image. The styles themselves are the same hand-written natural-language templates the pack is known for - prose that steers Z-Image (and its Qwen3-4B text encoder) toward a look.

How it works

The style dropdown pulls from the pack's style library, the gallery button opens the visual style browser, and string is your prompt. The node looks up the selected style's template and injects it around your text. The output STRING is the modified prompt - you can read exactly what changed, which makes it trivially debuggable and also lets you see the styled prompt even if you never use it.

The inputs that matter

• style (COMBO) - the visual style.
• gallery - visual style picker.
• string (STRING, multiline) - the prompt to modify.
• Output: STRING - the styled prompt, ready for your encoder of choice.

That's it. No CLIP, no conditioning output, no model math.

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.

The honest take

This node is the right call when the full Style & Prompt Encoder feels like too much - when you already have a conditioning setup you trust, or when you want the styled text visible and editable between style and encode. The trade-off is exactly the flip side: you now own the encoding step, so forgetting to hook the output into a text encoder produces a silent "nothing happened" instead of an error. If you're new to Z-Image and just want images, the all-in-one encoder is simpler. If you're assembling a modular graph, this is the piece that keeps style from boxing you in. The original v1 is deprecated, so if a workflow shows it, swap to this v2.