Z Image is a local text-to-image model designed for high-quality and diverse image generation in ComfyUI. It supports realistic photography, cinematic digital art, anime, and stylized illustration styles, all from text prompts.

Do I need a negative prompt for Z Image?

Yes. Z Image relies on negative prompts for optimal quality. Without a negative prompt, outputs often contain more unwanted artifacts, odd details, or composition issues. Always fill in the negative prompt field for best results.

What are the best sampling settings for Z Image?

Start with 30–50 steps and a CFG scale of 4.0–7.0. These are reliable starting points that you can adjust based on your GPU and the level of detail you want in the output.

Z Image ComfyUI Workflow: Complete Text-to-Image Guide (2026)

Getting Started with Z Image for Text-to-Image Generation

What you'll learn: How to install the Z Image model in ComfyUI, configure your workflow nodes, write effective prompts and negative prompts, set the best sampling parameters, and generate high-quality images across four different style categories.

Z Image is a local text-to-image model built for high-quality and diverse image generation inside ComfyUI. It supports a wide range of styles — from hyper-realistic photography to cinematic art, anime, and stylized illustrations — all from text prompts. If you're looking for a flexible, capable model to run locally, Z Image is a strong choice.

What is Z Image?

Z Image is the main model in the Z Image family. It's designed for quality and creative diversity — letting you create photos, digital art, anime, and illustrations from text alone. There's also a faster variant called Z Image Turbo, but the standard Z Image model prioritises image quality and creative flexibility.

🎨

Versatile styles

Handles realistic photos, anime, cinematic art, and stylized illustrations from a single model.

🖼️

Output variety

Generates more unique faces, lighting setups, and scene compositions with each run.

Sample Outputs from Z Image

Here are example results across the four main style categories. Each includes the full prompt so you can copy and use it directly in your own workflow.

Hyper-Realistic Photography

Z Image produces images that closely resemble real photographs — accurate skin texture, natural lighting, and true-to-life proportions.

Realistic Example 1 — Japanese woman in a purikura arcade booth

Z Image hyper-realistic output — young Japanese woman taking a selfie in a purikura photo booth with neon arcade backdrop and authentic lighting — Z Image — hyper-realistic photography, arcade selfie scene

Prompt used

Ultra-realistic professional photography of a young Japanese woman taking a playful mirror selfie inside a real purikura photo booth. She has tousled bangs, slightly smudged eyelin…

Realistic Example 2 — Italian fashion model on a rooftop terrace at golden hour

Z Image hyper-realistic output — 28-year-old Italian woman with auburn curly hair on a cinematic European rooftop terrace at golden hour — Z Image — editorial fashion photography, European rooftop scene

Prompt used

Ultra-realistic professional fashion photography of a 28-year-old Italian woman captured in a bold, confident moment, standing effortlessly at the center of a richly detailed Europ…

Realistic Example 3 — Mature woman, full-body studio portrait

Z Image hyper-realistic output — mature woman model full-body view with natural aging details, realistic skin texture, and shallow depth of field — Z Image — full-body portrait with realistic aging detail and natural lighting

Prompt used

Ultra-realistic professional photography of a mature woman model, full-body view, standing naturally, confident and elegant posture. Natural facial proportions, realistic aging de…

Cinematic Digital Art

Z Image excels at dramatic, film-quality digital art with layered lighting, moody colour grading, and cinematic composition.

Cinematic Example 1 — Sun-drenched lifestyle drama, Mediterranean coastal town

Z Image cinematic digital art output — 28-year-old American woman in a floral sundress at a Mediterranean coastal town, warm golden-hour cinematic lighting — Z Image — cinematic lifestyle drama, golden hour Mediterranean setting

Prompt used

Cinematic art depicting a radiant, sun-drenched film still from a modern lifestyle drama, featuring a strikingly attractive 28-year-old American woman as the central character. She…

Cinematic Example 2 — Rain-soaked European street, dramatic night scene

Z Image cinematic digital art output — Italian woman with auburn hair on a rain-soaked European street at night with neon reflections and dramatic film lighting — Z Image — cinematic film still, dramatic European night atmosphere

Prompt used

Cinematic art depicting a powerful, emotionally charged film still featuring a 28-year-old Italian woman as the central character, captured in a dramatic moment frozen in time. She…

Cinematic Example 3 — Action film still, urban rooftop stunt

Z Image cinematic digital art output — powerful woman of mixed Native American and African heritage mid-stunt on an urban rooftop with city skyline backdrop and dramatic natural lighting — Z Image — high-budget action film still, power and athleticism captured mid-stunt

Prompt used

Cinematic art depicting a breathtaking, high-impact film still frozen at the peak of a stunt, featuring a strikingly beautiful and powerful woman of mixed Native American and Afric…

Anime & Manga Styles

Z Image handles anime and manga-inspired art with dynamic composition, bold linework, and cinematic energy.

Anime Example 1 — Mid-air battle in a flooded cyber-shrine city

Z Image anime output — Japanese female fighter with silver-black hair performing a spinning kick mid-air in a flooded cyberpunk shrine city with neon lighting and motion blur — Z Image — next-gen anime action illustration, cyber-shrine battle scene

Prompt used

next-gen anime action illustration, Flux-Anime / modern anime diffusion style, ultra-dynamic cinematic frame, a 25-year-old Japanese female fighter with long silver-black hair in a…

Anime Example 2 — Rooftop parkour in cyberpunk Tokyo

Z Image anime output — Japanese woman with flowing ponytail leaping between neon-lit rooftops in a rain-soaked futuristic Tokyo cityscape — Z Image — cinematic anime action, cyberpunk Tokyo rooftop jump

Prompt used

Anime & manga cinematic action scene, a 25-year-old Japanese woman with long flowing ponytail hair, sharp expressive anime eyes, determined confident expression, performing a mid-a…

Anime Example 3 — Karate warrior vs demonic villain in a neon dojo city

Z Image anime output — Japanese woman warrior in modernized karate gi mid-strike against a four-armed demon villain with glowing eyes in a ruined neon dojo city — Z Image — manga cinematic fight scene, ultra-detailed martial arts action

Prompt used

Anime & manga cinematic illustration, ultra-detailed mid-fight scene, a 25-year-old Japanese woman warrior with long jet-black hair tied in a flowing high ponytail, strands whippin…

Stylized Illustrations

Z Image produces creative, editorial-grade stylized illustrations that blend surrealism, fashion, and bold graphic design.

Illustration Example 1 — Southeast Asian woman in a floating surreal marketplace

Z Image stylized illustration output — Southeast Asian woman in a futuristic high-fashion outfit reaching toward a floating holographic sphere in a surreal floating marketplace city — Z Image — editorial illustration meets surrealism, floating city fashion concept art

Prompt used

ultra-stylized digital illustration, modern fashion concept art fused with surrealism, vibrant and artistic composition, a 27-year-old Southeast Asian woman with asymmetrical paste…

Illustration Example 2 — Latin-American woman in a surreal floating cityscape

Z Image stylized illustration output — Latin-American woman with teal and midnight-blue gradient hair in a conceptual couture outfit against a surreal floating geometric cityscape — Z Image — ultra-stylized high-fashion editorial illustration, surreal environment

Prompt used

ultra-stylized illustration, high-fashion editorial art style, bold graphic shapes and expressive color theory, a stylish Latin-American woman around 27 years old, sharp jawline, c…

Illustration Example 3 — African-American woman in a futuristic urban surreal scene

Z Image stylized illustration output — African-American woman with platinum bob in futuristic couture holding a translucent geometric orb against a fragmented glass cityscape — Z Image — editorial concept art, futuristic fashion illustration with painterly textures

Prompt used

ultra-stylized illustration, editorial concept art meets fashion illustration, bold graphic design, a 30-year-old African-American woman with sleek platinum bob haircut, confident …

Using Z Image with ComfyUI

Setting up Z Image in ComfyUI takes about 10 minutes. Follow these steps in order.

Update ComfyUI

Before installing any new model, make sure ComfyUI and all your nodes are up to date:

Open ComfyUI Manager from inside ComfyUI.
Click "Update All" to update all installed custom nodes.
Restart ComfyUI after the update completes.

ComfyUI Manager showing the Update All button to update all custom nodes before installing Z Image — Click "Update All" in ComfyUI Manager before adding new models

Download Required Models

Download all three model files from Hugging Face and place them in the correct ComfyUI subfolders:

Text Encoder: qwen_3_4b.safetensors
Diffusion Model: z_image_bf16.safetensors
VAE: ae.safetensors

📂 ComfyUI/
└── 📂 models/
    ├── 📂 text_encoders/
    │   └── qwen_3_4b.safetensors
    ├── 📂 diffusion_models/
    │   └── z_image_bf16.safetensors
    └── 📂 vae/
        └── ae.safetensors

Download the Workflow

Download the ready-made Z Image workflow and drag it into your ComfyUI canvas to load it:

Download Z Image Workflow JSON

Setting Up Your Workflow

After loading the workflow, select the correct model files in each node:

Diffusion Modelz_image_bf16.safetensors

Text Encoderqwen_3_4b.safetensors

VAEae.safetensors

ComfyUI workflow nodes showing the diffusion model, text encoder, and VAE model files selected for Z Image — Select the correct model files in each node after loading the workflow

Set your output width and height in the resolution node:

ComfyUI resolution node showing width and height settings for Z Image generation — Set your output dimensions — Z Image works best between 512×512 and 2048×2048

⚠️ Note: Z Image works best with resolutions between 512×512 and 2048×2048 pixels in any aspect ratio. If you need a larger image, generate at a lower resolution first and then use an upscaler.

Writing Prompts and Negative Prompts

Enter your main prompt in the positive prompt node describing what you want to create. Always use a negative prompt — Z Image relies on it for optimal quality. Without a negative prompt, outputs frequently contain unwanted artifacts, odd proportions, or composition issues.

✓ With negative prompt

✗ Without negative prompt

Best Settings for Sampling

These are reliable starting settings for the KSampler node:

Steps30 – 50

CFG Scale4.0 – 7.0

ComfyUI KSampler node showing recommended steps and CFG scale settings for Z Image — Recommended KSampler settings — 30–50 steps, CFG 4.0–7.0

💡 Tip: Adjust these values based on your GPU. Higher steps increase detail but take longer. Lower CFG gives more creative freedom; higher CFG follows your prompt more strictly.

Running the Workflow

Once everything is configured, click Queue Prompt in ComfyUI (or press Ctrl+Enter / Cmd+Enter) to generate your image. The model will process your prompt and produce output based on your settings.

Tips and Troubleshooting

Blank or broken output images

Fix: Check that all model files are in the correct folder paths and the filenames match exactly. Refresh ComfyUI after placing new files.

Weird artifacts or odd details in the output

Fix: Add or improve your negative prompt — this is the most common cause of artifacts in Z Image. You can also try lowering the CFG scale or reducing the step count.

Resolution errors or out-of-memory crashes

Fix: Make sure your width × height stays between 512×512 and 2048×2048. If you need larger output, generate smaller and use an upscaler.

Missing nodes after loading the workflow

Fix: Open ComfyUI Manager, click "Update All", then restart ComfyUI. If a specific node is still missing, search for it in the Manager and install it manually.

Frequently Asked Questions

Z Image is a local text-to-image model built for ComfyUI. It's designed for quality and creative diversity, supporting realistic photography, cinematic digital art, anime, and stylized illustration styles — all from text prompts, without any external API.

You need three files: z_image_bf16.safetensors (diffusion model), qwen_3_4b.safetensors (text encoder), and ae.safetensors (VAE). All are available to download for free from Hugging Face.

Z Image works best between 512×512 and 2048×2048 pixels in any aspect ratio. For larger outputs, generate at a lower resolution first and then upscale using an upscaler workflow.

Yes — Z Image relies on negative prompts for optimal quality. Without a negative prompt, outputs often contain artifacts, odd proportions, or unusual details. Always fill in the negative prompt field for best results.

Start with 30–50 steps and a CFG scale between 4.0 and 7.0. These are reliable starting points. Adjust based on your GPU — higher steps increase detail but take longer, and lower CFG gives the model more creative freedom.