Z-Image Turbo ControlNet Union 2.1: Complete Multi-Control Workflow Guide

Abstract: Z-Image Turbo ControlNet Union 2.1 is a unified control network model open-sourced by Alibaba's PAI team, supporting depth, edge, normal, semantic segmentation, grayscale, and multiple other control conditions. This article takes you from basic concepts to advanced multi-control workflows, teaching you Z-Image's multi-control point combination generation techniques step by step.

Introduction

ControlNet is the core technology for achieving precise spatial control in AI image generation. In early 2026, Alibaba's PAI team officially open-sourced Z-Image-Turbo-Fun-ControlNet-Union 2.1, a unified ControlNet model specifically trained for Z-Image Turbo.

Unlike traditional approaches requiring separate ControlNet models for each control condition, ControlNet Union consolidates all control conditions into a single model file, significantly simplifying workflows and improving flexibility.

1. What is ControlNet Union?

1.1 Basic Concepts

ControlNet Union is a unified ControlNet model supporting these control conditions:

Control Type	Description	Use Cases
Depth	Spatial structure from depth maps	Architecture, scene composition
Canny	Contour control via Canny edge detection	Line art coloring, sketch refinement
Normal	Surface orientation from normal maps	3D rendering, material replacement
Seg	Region control via semantic segmentation	Scene editing, object replacement
Gray	Light/dark control from grayscale	B&W to color, tone control
Pose	Human pose control	Portrait generation, action control
Tile	Local control via tile maps	Image expansion, detail enhancement

1.2 Technical Background

Model Size: Single file ~2GB (smaller than multiple independent ControlNets)
Training Data: 1 million high-quality images
Training Steps: 10,000 steps
Architecture: ControlNet added on 6 DiT blocks of Z-Image Turbo
Download: HuggingFace - alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union

2. ComfyUI Environment Setup

2.1 Install ComfyUI and Extensions

# Install ComfyUI
git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
pip install -r requirements.txt

# Install ComfyUI Manager
cd custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Manager.git

2.2 Download Model Files

Z-Image Turbo Main Model (place in models/unet/):

z-image-turbo-fp8-e4m3fn.safetensors

ControlNet Union 2.1 (place in models/controlnet/):

Z-Image-Turbo-Fun-Controlnet-Union.safetensors

CLIP Text Encoder (place in models/clip/):

qwen3-4b-fp8-scaled.safetensors

2.3 Minimum Hardware Requirements

Configuration	VRAM	Notes
Entry Level	8GB	FP8 quantization + basic resolution
Recommended	12GB	FP8 + high resolution + multi-ControlNet
Ideal	16GB+	BF16 precision + high resolution

3. Basic Workflows

3.1 Single Control Condition Workflow

The most basic ControlNet workflow includes these nodes:

Load Checkpoint (Z-Image-Turbo)
  → ControlNet Loader (Union 2.1)
  → ControlNet Preprocessor (select control type)
  → Apply ControlNet
  → KSampler (8 steps)
  → Save Image

Canny Edge Detection Example:

Control type: canny
ControlNet strength: 0.8-1.0
Use case: Line art coloring, sketch refinement

3.2 Depth Map Control Workflow

Depth control is one of the most commonly used scenarios:

Input: Depth map (can be auto-generated from RGB)
Control type: depth
Application: Change style while maintaining spatial structure

# Key ComfyUI workflow parameters
{
    "control_net_name": "Z-Image-Turbo-Fun-Controlnet-Union.safetensors",
    "control_net_type": "depth",
    "strength": 0.9,
    "start_percent": 0.0,
    "end_percent": 1.0,
    "num_inference_steps": 8,
    "guidance_scale": 7.5
}

4. Multi-Control Combination Workflows (Core Feature)

The biggest advantage of ControlNet Union 2.1 is simultaneously using multiple control conditions. This is the core content of this article.

4.1 Depth + Canny Combination

Application: Precisely control spatial structure while maintaining contour details.

ControlNet 1: Depth (strength 0.8)
ControlNet 2: Canny (strength 0.6)
  ↓ Combined application
KSampler (8 steps)
  ↓
Output: Structurally precise + detail-rich images

Tips:

Depth controls spatial structure (front, higher strength)
Canny controls edge details (back, lower strength)
Total strength of both ControlNets: 1.2-1.4

4.2 Semantic Segmentation + Depth Combination

Application: Scene editing, object replacement, style transfer.

ControlNet 1: Seg (semantic segmentation, strength 0.7)
  → Controls what content is in which region
ControlNet 2: Depth (strength 0.7)
  → Controls spatial structure
  ↓ Combined application
Prompt: "Add a running hound dog on the grass"
  ↓
Output: Semantically correct scene editing results

4.3 Pose + Canny Combination

Application: Character generation, outfit changes, action control.

ControlNet 1: Pose (strength 0.9)
  → Controls character pose
ControlNet 2: Canny (strength 0.5)
  → Controls outfit/appearance contours
  ↓ Combined application
Prompt: "Business person in red suit, standing in office"
  ↓
Output: Accurate pose + outfit matching description

5. Advanced Techniques

5.1 Multi-Resolution Strategy

# Two-stage high-resolution strategy
Stage 1: 512×512 + ControlNet (structure control)
  → Use Hires.Fix / Latent Upscale
Stage 2: 1024×1024 (no ControlNet, detail enhancement)
  → Use Z-Image Turbo native 8-step inference

5.2 Grayscale to Color Workflow

ControlNet Union's Gray mode excels at coloring black-and-white photos:

Input black-and-white photo
Control type: gray
Prompt describes target color style
Output: Colorized image

Tips:

Gray mode strength: 0.7-0.8
Combine with depth control to maintain structure
Suitable for old photo restoration, sketch coloring

5.3 ControlNet Strength Tuning Tips

Scenario	Recommended Strength	Notes
Structure (Depth)	0.8-1.0	High strength for spatial structure
Edge (Canny)	0.5-0.8	Medium strength for detail flexibility
Semantic (Seg)	0.6-0.9	Adjust based on edit scope
Pose	0.8-1.0	High strength for pose accuracy
Grayscale (Gray)	0.6-0.8	Balance light/dark and color
Multi-control combo	0.4-0.7 each	Total ≤ 1.5

6. Common Application Scenarios

6.1 Architectural Design

ControlNet combo: Depth + Normal
Prompt: "Modern glass facade building, sunset light, reflecting sky"
Effect: Maintain building structure, change materials and lighting

6.2 E-commerce Product Photography

ControlNet combo: Canny + Tile
Prompt: "Professional product photography, marble surface, soft studio lighting, white background"
Effect: Generate professional product images from product sketches

6.3 Character Design

ControlNet combo: Pose + Canny + Seg
Prompt: "Cyberpunk-styled character, neon-lit city background"
Effect: Precisely control character pose, outfit contours, and scene elements

6.4 Image Expansion (Uncrop)

ControlNet combo: Tile + Depth
Prompt: "Expand the frame, show a wider city skyline"
Effect: Intelligently expand image boundaries, maintaining style consistency

7. Troubleshooting

7.1 Insufficient VRAM

# Solution 1: Use FP8 quantized models
# Solution 2: Reduce resolution
# Solution 3: Enable --lowvram mode
python main.py --lowvram

7.2 Weak Control Effect

Check if ControlNet strength is too low
Confirm correct control type selection
Verify preprocessing map quality (depth, edges, etc.)
Try increasing start_percent (from 0.0 to 0.1-0.2)

7.3 Over-Control Causing Stiff Images

Reduce ControlNet strength
Set end_percent < 1.0 (let model be free in final steps)
Reduce the number of simultaneously used ControlNets

8. Complete Workflow JSON Template

Key node configuration for ComfyUI multi-control combination workflow:

{
  "controlnet_depth": {
    "model": "Z-Image-Turbo-Fun-Controlnet-Union.safetensors",
    "type": "depth",
    "strength": 0.8,
    "start_percent": 0.0,
    "end_percent": 0.9
  },
  "controlnet_canny": {
    "model": "Z-Image-Turbo-Fun-Controlnet-Union.safetensors",
    "type": "canny",
    "strength": 0.6,
    "start_percent": 0.0,
    "end_percent": 0.8
  },
  "sampler": {
    "steps": 8,
    "cfg": 7.5,
    "scheduler": "euler",
    "model": "z-image-turbo-fp8-e4m3fn.safetensors"
  }
}

9. Summary

Z-Image Turbo ControlNet Union 2.1 is one of the most powerful control tools in the Z-Image ecosystem. Its unified architecture, multi-control combination capabilities, and open-source nature make it the preferred choice for professional image generation workflows.

Key Takeaways:

Unified Model: One model supports all control types
Multi-Control Combination: Use multiple ControlNets simultaneously for precise control
Strength Tuning: Properly distribute strength across ControlNets
Scenario Adaptation: Select different control combinations for different scenarios
FP8 Deployment: Runs even in low VRAM environments

Mastering ControlNet Union 2.1 means mastering Z-Image Turbo's core competitive advantage.

Related Articles:

ZI-072: Z-Image ComfyUI Power Nodes Advanced Workflow
ZI-058: Z-Image ControlNet Union 2.1 Multi-Control Practical Guide
ZI-080: Z-Image Batch Generation and Performance Tuning

Z-Image Turbo ControlNet Union 2.1: Complete Multi-Control Workflow Guide

Table of Contents

Z-Image Turbo ControlNet Union 2.1: Complete Multi-Control Workflow Guide

Introduction

1. What is ControlNet Union?

1.1 Basic Concepts

1.2 Technical Background

2. ComfyUI Environment Setup

2.1 Install ComfyUI and Extensions

2.2 Download Model Files

2.3 Minimum Hardware Requirements

3. Basic Workflows

3.1 Single Control Condition Workflow

3.2 Depth Map Control Workflow

4. Multi-Control Combination Workflows (Core Feature)

4.1 Depth + Canny Combination

4.2 Semantic Segmentation + Depth Combination

4.3 Pose + Canny Combination

5. Advanced Techniques

5.1 Multi-Resolution Strategy

5.2 Grayscale to Color Workflow

5.3 ControlNet Strength Tuning Tips

6. Common Application Scenarios

6.1 Architectural Design

6.2 E-commerce Product Photography

6.3 Character Design

6.4 Image Expansion (Uncrop)

7. Troubleshooting

7.1 Insufficient VRAM

7.2 Weak Control Effect

7.3 Over-Control Causing Stiff Images

8. Complete Workflow JSON Template

9. Summary