⚗️🔬| Style Library v1.0

Z-Image Turbo's weakness was never photorealism. It's style. The model will give you a gorgeous, realistic render almost on autopilot, but push it toward a niche look - lo-fi phone photography, a 1970s family album, a stylized illustration - and you're suddenly hand-writing prompts that fight its defaults. Style Library v1.0 is the same author's answer, and it's the same idea as its Palette sibling: pick from a gallery, stop hand-wrangling prompts.

It comes from ComfyUI-ZImagePowerNodes, the pack by Martin Rizzo (of Amazing Z-Image Workflow fame), built specifically for the Z-Image / Z-Image Turbo models. Like the palette picker, this node is pure selection - it doesn't encode anything itself. It hands a chosen style object to the StylePromptEncoderX22Advanced node, which is the one that actually builds the conditioning.

How it works

Version 1.0 ships 123 predefined styles - 54 photographic, 36 illustration, 33 "wild" - each defined as a hand-tuned prompt template in plain-text files (styles/styles-v10-*.sty.txt). Open one and you'll see why they work. They're not "cinematic, moody lighting, 8k" noise. They're structured blocks:

>>>Phone Photo
YOUR CONTEXT:
Your photographs has android phone cam-quality.
Your photographs exhibit surprising compositions, sharp complex backgrounds, natural lighting, and candid moments...
YOUR PHOTO:
{$@}

The {$@} is where your actual prompt gets dropped in. And notice the vocabulary: these styles name the source - phone camera, selfie, family album - rather than reaching for mood words. That's not an accident. The Z-Image community spent months discovering that the model breaks out of its default "influencer" look when you prompt physical cameras and film stocks instead of quality adjectives. These templates are that finding, packaged.

Because styles are versioned (v1.0 here, newer ones in the pack), the author can iterate on the library without breaking workflows that reference the old set.

The input and output

One input, one output, both worth knowing:

• style (type ZIPN_PREDEFINED_STYLE) - opens a searchable gallery dialog with thumbnail samples for each style. Browsing 123 styles in a dropdown would be miserable; the gallery makes it actually pleasant.
• Output STYLE (type ZIPN_CUSTOM_STYLE) - wire this into the style input of StylePromptEncoderX22Advanced. It's a pack-internal type, so a generic node won't know what to do with it.

If your subject and composition matter more than the look, this is the node you want: the style only shifts the artistic direction while your prompt keeps the subject intact.

Installing it

Same pack as the whole Z-Image Power Nodes suite:

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

Restart, or use ComfyUI Manager (search "Z-Image Power Nodes") if you prefer clicking. There are no Python dependencies to install - the pack is pure ComfyUI-API code. What you do need are the Z-Image models themselves: the Turbo diffusion model, the Qwen3-4B text encoder, and the Flux 1 VAE (ae.safetensors, 335 MB). GGUF checkpoints additionally require the ComfyUI-GGUF pack.

The honest caveats

This node sits in the ⚡Z-Image/__experimental category, and its own description warns that experimental nodes may change or be removed without notice. It's brand new with zero search traffic behind it, so consider it a preview rather than a settled API. The styles themselves are the real value - and if the stable Style Prompt Encoder in the same pack already covers what you need, you don't strictly need this experimental selector at all. You're here because you want the gallery and the newer style set. Fair enough - just don't build a production workflow around an experimental picker.