Generating images with ComfyUI and Z Image Turbo

Overview

ComfyUI is a powerful, node-based interface for Stable Diffusion and other diffusion models. Unlike traditional text-to-image interfaces with simple prompt boxes, ComfyUI exposes the entire image generation pipeline as a visual graph, giving you fine-grained control over every step from text encoding to latent space manipulation to final decoding.

This tutorial teaches you how to use ComfyUI with the Z Image Turbo model on your GPU to generate high-quality AI images.

What You’ll Learn

• How to launch ComfyUI and load the Z-Image Turbo template
• Understanding diffusion pipeline components
• Generating images and tuning generation parameters
• Saving and sharing workflows

Installing Dependencies

sudo apt update
sudo apt install -y python3-venv
python3 -m venv llm-env --system-site-packages
source llm-env/bin/activate

sudo apt update
sudo apt install -y python3-venv
python3 -m venv llm-env
source llm-env/bin/activate

git clone https://github.com/Comfy-Org/ComfyUI.git

git checkout v0.17.2

pip install -r requirements.txt

sudo apt update
sudo apt install -y python3-venv
python3 -m venv llm-env --system-site-packages
source llm-env/bin/activate

sudo apt update
sudo apt install -y python3-venv
python3 -m venv llm-env
source llm-env/bin/activate

git clone https://github.com/Comfy-Org/ComfyUI.git

git checkout v0.17.2

pip install -r requirements.txt

sudo apt update
wget https://repo.radeon.com/amdgpu-install/7.2.1/ubuntu/noble/amdgpu-install_7.2.1.70201-1_all.deb
sudo apt install ./amdgpu-install_7.2.1.70201-1_all.deb
sudo amdgpu-install -y --usecase=rocm --no-dkms

sudo usermod -aG render,video $USER

sudo reboot

ls -l /opt/rocm
ls -l /opt/rocm/lib/libroctx64.so*

# Check ROCm device files (Device files owned by the render group should be visible)
ls -l /dev/kfd
ls -l /dev/dri/renderD*

# Check user groups ($USER should be listed in both render and video)
id
groups

# Check ROCm with rocminfo ('Permission denied' error should NOT be seen)
rocminfo | sed -n '1,120p'

# Check installed ROCm version
cat /opt/rocm/.info/version

python -m pip install --upgrade pip
python -m pip install --force-reinstall --no-cache-dir --index-url https://repo.amd.com/rocm/whl/gfx1151/ torch torchvision torchaudio

python -m pip install --upgrade pip
python -m pip install --force-reinstall --no-cache-dir --index-url https://repo.amd.com/rocm/whl/gfx1152/ torch torchvision torchaudio

python -m pip install --upgrade pip
python -m pip install --force-reinstall --no-cache-dir --index-url https://repo.amd.com/rocm/whl/gfx1150/ torch torchvision torchaudio

Launching ComfyUI

Finding the Z-Image Turbo Template

Before generating images, you need to load the Z-Image Turbo template. Here’s how to find it:

1. Look at the far left edge of the screen—there’s a vertical toolbar running from top to bottom on the leftmost side of the app.

2. Find the folder icon—in that left toolbar, look for an icon that looks like a folder. When you hover over it, it’s labeled “Templates.”

3. Click the folder icon—this opens the Templates panel.

4. Search for “Z-Image Turbo”—use the search bar or scroll through the available templates to find the Z-Image Turbo Text To Image workflow, then click to load it.

Downloading Models

Understanding the Interface

When the Z-Image Turbo template loads, you’ll see a canvas with 2 main nodes. The first node is called ‘Text to Image (Z-Image-Turbo), and the second node is for viewing the image.

On the Z-Image node, click the top right button to expand the Node and see the subgraph.

Pipeline Components

The Z-Image Turbo workflow uses four key model components that work together:

Each node in the workflow corresponds to one of these components. Data flows left-to-right: text → embeddings → guided denoising → latents → final image.

Generating Your First Image

The Z-Image Turbo model is already loaded. To generate an image:

1. Enter your prompt in the main Z-Image Node. Be descriptive. Here is an example:

A photorealistic red fox sitting in a snowy forest clearing,
morning light filtering through pine trees,
detailed fur texture, bokeh background

2. (Optional): Confirm or tweak any other specific settings within the subgraph.
3. Click the blue “Run Workflow” in the right corner (or press Ctrl+Enter)
4. Watch the nodes highlight as each step executes

The entire workflow execution should complete in less than 30 seconds. Your generated image appears in the Save Image node and is saved to the output/ folder.

Adjusting Generation Parameters

KSampler Settings

The KSampler node controls the core diffusion process:

Image Size

To adjust output dimensions, find the Empty Latent Image node and modify width and height. Keep dimensions at or below 1024 pixels on the longest side for optimal quality.

ModelSamplingAuraFlow

The ModelSamplingAuraFlow node is a specialized sampling modifier that adjusts how the diffusion process handles noise scheduling. You’ll see this node connected to the model output in the Z-Image Turbo workflow.

When to adjust shift:

• Lower values (1.0–2.0): Faster convergence, good for simple compositions
• Higher values (3.0–4.0): More gradual refinement, can improve fine details in complex scenes

The AuraFlow sampling method is specifically designed for flow-matching models like Z-Image Turbo, ensuring proper noise distribution throughout the generation process.

Working with Workflows

Saving Workflows

Click the Save button in the menu to export your workflow as a JSON file. This captures:

• All nodes and their parameters
• All connections between nodes
• Current prompt text

Loading Workflows

Drag a workflow JSON file onto the canvas, or use Load from the menu. The Z-Image Turbo workflow you see by default is loaded from a saved workflow file.

Sharing Workflows

Workflows are self-contained—share the JSON file with colleagues, and they can reproduce your exact setup. This makes ComfyUI excellent for collaborative experimentation.

Next Steps

• Explore LoRA nodes: Apply style or subject adapters without retraining
• Add negative prompts: Connect a second CLIP Text Encode node to the negative conditioning input of KSampler to guide the model away from unwanted features like blur, artifacts, or watermarks
• Build custom workflows: Chain multiple generations, add upscaling, or create image variations
• Browse community workflows: ComfyUI Examples has many ready-to-use workflows

ComfyUI’s strength is experimentation: connect nodes differently, adjust parameters, and observe how each change affects the output. This hands-on exploration builds intuition for how diffusion models work.

For more information, check out the ComfyUI Documentation.