Z-Image ComfyUI Power Nodes Advanced Workflow: Complete Custom Node Configuration Guide

Published: June 1, 2026
Keywords: z-image comfyui power nodes, z-image custom nodes, ComfyUI Z-Image workflow
Reading time: 15 minutes

---

Introduction

ComfyUI has become the standard tool in AI image generation, beloved by developers and creators alike for its node-based workflow architecture. For Z-Image users, ComfyUI-ZImagePowerNodes (abbreviated as Power Nodes) is the key extension that unlocks the full potential of the Z-Image model.

Created by community developer martin-rizzo, Power Nodes provides a suite of custom nodes specifically optimized for Z-Image. These nodes solve multiple pain points in native ComfyUI's Z-Image integration: text encoding formats, LoRA loading optimization, CFG scheduling strategies, and more.

This guide takes you from zero to mastery — covering Power Nodes installation, configuration, and deep dives into every core node's usage and best practices.

---

Installation and Configuration

Prerequisites

• ComfyUI installed and running properly
• Z-Image Turbo or Base model downloaded
• Python 3.10+ environment

Install via ComfyUI Manager (Recommended)

1. Open ComfyUI and click the Manager button
2. Click Custom Nodes Manager
3. Search for ComfyUI-ZImagePowerNodes
4. Click Install
5. Restart ComfyUI

Manual Installation

cd ComfyUI/custom_nodes
git clone https://github.com/martin-rizzo/ComfyUI-ZImagePowerNodes.git
cd ComfyUI-ZImagePowerNodes
pip install -r requirements.txt

Model Files Preparation

Power Nodes requires these model files:

File Type	Filename	Download Source
Z-Image Turbo	z_image_turbo.safetensors	HuggingFace
Z-Image Base	z_image_base.safetensors	HuggingFace
Text Encoder	Qwen3-4B related models	HuggingFace
VAE	vae.safetensors	Model repository

---

Core Nodes Explained

ZImageTextEncoder — Text Encoding Node

This is the most critical node in Power Nodes, responsible for converting natural language prompts into Z-Image's required encoding format.

Input Parameters:

Parameter	Type	Description
prompt	STRING	Natural language prompt
clip	CLIP	Connected from CLIPLoader
cfg	FLOAT	Guidance scale (recommended 1.0-2.0)

Key Features:

• Automatically formats prompts into Qwen3-4B chat template format
• Supports think block for enhanced reasoning
• Built-in system prompt template optimization

ZImageTurnBuilder — Multi-turn Conversation Builder

Used to construct complex conversational prompts, supporting multi-turn interaction context understanding.

Use Cases:

• Multi-round descriptions requiring precise image detail control
• Step-by-step construction of complex scenes
• Combining system prompts with user prompts

Typical Workflow:

[CLIPLoader] → [ZImageTurnBuilder] → [ZImageTextEncoder] → [KSampler] → [VAEDecode]

ZImageLoRA — LoRA Loading Node

This is the biggest pain point Power Nodes solves. The native ComfyUI LoRA node frequently causes lighting anomalies and color distortion when working with Z-Image.

Differences from Standard LoRA Node:

Feature	Standard LoRA Node	ZImageLoRA Node
Lighting Fix	❌	✅
Color Correction	❌	✅
Adaptive Rank	❌	✅
Linear/LoRA Toggle	❌	✅

Recommended Settings:

• LoRA strength: 0.6-0.8 (too high causes overfitting)
• Use linear rank (recommended 64-128)

ZImageSampler — Sampler Node

A sampler specifically optimized for Z-Image, supporting:

• Turbo mode: 4-8 step fast sampling
• Base mode: 20-50 step high-quality sampling
• Adaptive steps: Automatically adjusts based on CFG and seed

---

Building Complete Workflows

Basic Workflow

[Load Checkpoint] → [ZImageTextEncoder] → [ZImageSampler] → [VAEDecode] → [Save Image]

Step-by-step:

1. Load Checkpoint: Load Z-Image Turbo model
2. ZImageTextEncoder: Input prompt, connect CLIP
3. ZImageSampler: Set steps (Turbo=8, Base=28)
4. VAEDecode: Decode latent space to pixel space
5. Save Image: Save output

Advanced Workflow (with LoRA + ControlNet)

[Load Checkpoint]
    → [ZImageLoRA] → [ZImageTextEncoder]
[ControlNet Loader] → [ControlNet Apply] → [ZImageSampler] → [VAEDecode] → [Save Image]

Batch Generation Workflow

[Load Checkpoint]
    → [ZImageTextEncoder] → [ZImageSampler] → [VAEDecode]
[Batch Prompt Loader] → [ZImageTextEncoder]
[Image Saver (Batch)]

---

Prompt Template System

Power Nodes includes a powerful prompt template system located at:

custom_nodes/comfyui-z-image/nodes/templates/z_image/

Template Format

[System Prompt]
{system_prompt}

[Think Block]
<think>
{thinking_content}
</think>

[User Prompt]
{assistant_content}

Common Templates

Photography Style Template:

You are a professional photographer. Create a photorealistic image with
natural lighting and realistic textures.

Illustration Style Template:

You are an illustration artist. Create a stylized artwork with clean lines
and vibrant colors.

Product Photography Template:

You are a product photographer. Create a studio-quality product shot with
clean background and professional lighting.

Template Usage Tips

1. Customize system prompts: Adjust system_prompt based on output style
2. Use Think Blocks: Add reasoning step descriptions in think tags to improve output quality
3. Character counting: Power Nodes displays character counts and token estimates for system, user, and think sections
4. Direct mode: For simple prompts, use mode: direct to skip template formatting

---

Advanced Tuning Techniques

CFG Scheduling

Z-Image is very sensitive to CFG values:

CFG Value	Effect	Recommended For
0.5-1.0	Lower prompt adherence, more natural look	Artistic creation, abstract style
1.0-1.5	Balanced prompt adherence	General use (recommended)
1.5-2.0	High prompt adherence	Precise text rendering, product photography
2.0+	Too high, may cause oversaturation	Not recommended

Seed Strategies

Strategy	Description	Use Case
Fixed seed	Reproducible results	Debugging, A/B testing
Random seed	Diversity exploration	Creative exploration
Seed offset	Slight variations on base composition	Batch variant generation

Resolution Settings

Resolution	VRAM Requirement	Use Case
512×512	~4GB	Quick preview, thumbnails
768×768	~6GB	Social media graphics
1024×1024	~8GB	Standard output
1536×1024	~12GB	Wide-format design
2048×2048	~16GB+	HD output (requires tiling)

---

Troubleshooting Common Errors

Error 1: Lighting Anomalies

Symptoms: Generated images show unnatural lighting
Cause: Using standard LoRA node instead of ZImageLoRA
Fix: Replace with ZImageLoRA node, adjust strength to 0.6-0.8

Error 2: Text Rendering Failure

Symptoms: Generated text is garbled or unreadable
Cause: CFG too low or incorrect text encoding format
Fix:

• Increase CFG to 1.5-2.0
• Verify ZImageTextEncoder is properly connected
• Check that text in prompt is wrapped in quotes

Error 3: OOM (Out of Memory)

Symptoms: CUDA out of memory error during execution
Cause: Resolution too high or insufficient VRAM
Fix:

• Lower resolution to 768×768
• Use Turbo mode to reduce steps
• Enable CPU offload mode

Error 4: Template Format Error

Symptoms: Power Nodes console shows template format warnings
Cause: Prompt doesn't match Qwen3-4B chat template
Fix:

• Use ZImageTextEncoder for automatic formatting
• Or manually follow the system → think → user template structure

---

Performance Optimization

Speed Tips

1. Use Turbo mode: 4-8 steps vs 20-50 steps
2. FP16 precision: Set dtype: float16 in extra_model_commands.yaml
3. GPU optimization: Ensure CUDA version matches GPU driver
4. Batch processing: Use batch workflows to reduce node switching overhead

Memory Management

# Add these to ComfyUI launch parameters
--lowvram  # When VRAM < 8GB
--cpu      # When no GPU available (very slow)

---

Workflow Templates

Power Nodes provides pre-configured workflow templates available in ComfyUI's Workflow Templates:

• Basic Generation: z-image-basic.json
• LoRA Fine-tuning: z-image-lora.json
• ControlNet: z-image-controlnet.json
• Batch Production: z-image-batch.json

---

Conclusion

ComfyUI-ZImagePowerNodes is currently the most comprehensive Z-Image ComfyUI integration available. By properly configuring and using these nodes, you can fully leverage Z-Image's strengths in text rendering, image quality, and generation speed.

For users new to Power Nodes, start with the basic workflow and gradually add advanced features like LoRA and ControlNet. Once you master the template system, you'll be able to create highly customized image generation pipelines.

---

Related Articles:

• Z-Image Bilingual Text Rendering Complete Guide
• Z-Image Prompt Engineering Complete Guide
• Z-Image ControlNet Union 2.1 Multi-Control Practice

Tags: #Z-Image #ComfyUI #PowerNodes #Workflow #CustomNodes #AIGeneration