Can I use negative prompts with z-image-turbo?

No. z-image-turbo operates at CFG=1.0, so negative prompts have no effect. To improve image quality, optimize your positive prompt instead.

What is the optimal number of steps for z-image-turbo?

8 steps is optimal. Increasing steps adds only marginal detail and is not worth the speed trade-off.

Can I use this workflow on ConoHa AI Canvas?

Yes, you can import and use this workflow directly in ConoHa AI Canvas's ComfyUI feature.

z-image-turbo ComfyUI Workflow Distribution | Pre-Configured Optimal Settings

We’re distributing a workflow for using z-image-turbo with ComfyUI. Optimal parameters are pre-configured, so download it and start using it immediately.

Workflow Features

Settings optimized for z-image-turbo (8 steps, euler, CFG=1.0)
Fast generation in 8 steps
Just enter a prompt and go

How to Install

Method 1: Import a JSON File

Download the workflow JSON below
Open ComfyUI
Drag and drop the JSON file onto the screen (or “Load” from the menu)

Download: Coming soon

Method 2: Drag & Drop the Workflow Image

ComfyUI can read workflow metadata embedded in PNG images.

Download the workflow image below
Drag and drop it onto the ComfyUI screen

Download: Coming soon

If you don’t have a ComfyUI environment yet, see the ConoHa AI Canvas Getting Started Guide. You can use ComfyUI from your browser.

Workflow Structure

This workflow consists of the following nodes.

Node List

Node	Role	Setting
CheckpointLoader	Load the z-image-turbo model	z-image-turbo
CLIPTextEncode (positive)	Prompt input	(Write your prompt here)
CLIPTextEncode (negative)	Negative prompt	Leave blank (see note below)
KSampler	Sampling	steps=8, sampler=euler, scheduler=ddim_uniform
VAEDecode	Image decoding	—
SaveImage	Save image	—

Workflow Diagram

CheckpointLoader
  ├── MODEL → KSampler
  ├── CLIP → CLIPTextEncode (positive) → KSampler (positive)
  ├── CLIP → CLIPTextEncode (negative) → KSampler (negative)
  └── VAE → VAEDecode → SaveImage
                ↑
            KSampler (output)

About Negative Prompts

Negative prompts do not work in z-image-turbo.

z-image-turbo is a distilled model operating at CFG=1.0, so the negative prompt mechanism (Classifier-Free Guidance) does not function. The CLIPTextEncode (negative) node can be left blank with no issues.

For improving image quality, optimizing the positive prompt is effective rather than negative prompts. See Prompt Best Practices for details.

Technical note: ComfyUI’s KSampler node requires a negative input, so an empty CLIPTextEncode node is connected to it. For more details on how negative prompts work, see Negative Prompt Guide.

KSampler Settings Explained

Parameter	Value	Explanation
steps	8	z-image-turbo is designed to converge with few steps. Increasing this barely changes quality
sampler_name	euler	The simplest and fastest sampler. Works well with z-image-turbo
scheduler	ddim_uniform	Divides steps evenly. Optimal for the short 8-step inference
cfg	1.0	z-image-turbo is designed to operate at CFG=1.0. Negative prompts do not function at this setting
denoise	1.0	1.0 for txt2img (text-to-image generation)

What Happens If You Change Parameters?

Increase steps (12–20) — Slightly more detail, but speed is sacrificed. 8 steps is sufficient as a baseline.
Change sampler — Other samplers like dpmpp_2m can work, but euler offers the best speed-quality balance.
Change cfg — Keep at 1.0 for z-image-turbo. Changing it degrades image quality.

Usage Tips

Writing Your Prompt

Enter your prompt in the “CLIPTextEncode (positive)” node in the workflow.

Tips for writing prompts are covered in detail in Prompt Basics, but the basics are:

Write the subject first
Supplementary details in the middle (clothing, pose, etc.)
Quality instructions at the end (camera settings, etc.)

Example Usage

portrait of a beautiful Japanese woman in her 20s, long black hair, wearing a light blue summer dress, standing in a sunflower field, golden hour lighting, shallow depth of field, 85mm lens, professional photography

Changing Image Size

The default is 1024x1024. To change it, modify the values in the “EmptyLatentImage” node.

Use Case	Size	Ratio
Square	1024x1024	1:1
Landscape	1280x720	16:9
Portrait	720x1280	9:16
Social media	1080x1080	1:1

Troubleshooting

Workflow Won’t Load

Your ComfyUI version may be outdated. Update to the latest version.
Check that the z-image-turbo model file is placed in the Checkpoints folder.

Image Quality Is Low

Check that steps is set to 8 (extremely few steps will degrade quality)
Check that the image size isn’t too small (512x512 minimum recommended)
Review your positive prompt → Prompt Best Practices

Generation Is Taking Too Long

z-image-turbo only needs 8 steps. Check if steps is unnecessarily high.
If VRAM is insufficient, reduce the image size.

Summary

Distributed in 2 formats: workflow JSON and drag & drop image
8 steps, euler, cfg=1.0 are the optimal settings for z-image-turbo
Negative prompts don’t work (due to CFG=1.0). Improve image quality by optimizing the positive prompt.

Next Steps

Learn how to write prompts → Prompt Basics
Verified prompt knowledge → Prompt Best Practices
Want to see examples → Prompt Examples
Want to set up from scratch → ConoHa AI Canvas Getting Started Guide
Want to automate with API → RunPod Serverless Guide