Z-Image General — ComfyUI workflow

9Graph nodes

8Node classes

4Inputs

3Models

Tags: flagshipz-imagegeneral2026-sota

Inputs (4)#

The typed parameter surface callers bind when they request this workflow. Enum options and numeric bounds are the values the workflow document declares.

promptstringdefault a serene meditation teacher with kind eyes

Subject description in natural prose

aspectenumdefault portrait_896x1152

Output aspect preset

options: portrait_896x1152square_1024landscape_1152x896

stepsintegerdefault 8min 4max 12

Turbo-distilled sampling steps (8 is the reference)

seedintegerdefault -1

Random seed (-1 for random)

ComfyUI node graph (9)#

The executable ComfyUI prompt graph: 9 nodes across 8 distinct node classes, wired by 9 data dependencies. Nodes tinted green come from a custom node pack this workflow declares; the rest are ComfyUI core / baked-community classes.

flowchart TD n0["1: UNETLoader"] n1["2: CLIPLoader"] n2["3: VAELoader"] n3["4: ModelSamplingAuraFlow"] n4["5: CLIPTextEncode"] n5["6: CLIPTextEncode"] n6["7: EmptySD3LatentImage"] n7["8: KSampler"] n8["9: VAEDecode"] n0 -->|model| n3 n1 -->|clip| n4 n1 -->|clip| n5 n3 -->|model| n7 n4 -->|positive| n7 n5 -->|negative| n7 n6 -->|latent_image| n7 n7 -->|samples| n8 n2 -->|vae| n8

Data-flow DAG — scroll to zoom, drag to pan.

Nodes (9)#

1UNETLoadercore

unet_name = z_image_turbo_bf16.safetensorsweight_dtype = default

outputs: MODEL

2CLIPLoadercore

clip_name = qwen_3_4b.safetensorstype = lumina2

outputs: CLIP

3VAELoadercore

vae_name = z_image_ae.safetensors

outputs: VAE

4ModelSamplingAuraFlowcore

model = ◂ node 1 · out[0]shift = 3.0

outputs: MODEL

5CLIPTextEncodecore

text = {{constructed_prompt}} tmplclip = ◂ node 2 · out[0]

outputs: CONDITIONING

6CLIPTextEncodecore

text = clip = ◂ node 2 · out[0]

outputs: CONDITIONING

7EmptySD3LatentImagecore

width = {{aspect_map[aspect].width}} tmplheight = {{aspect_map[aspect].height}} tmplbatch_size = 1

outputs: LATENT

8KSamplercore

model = ◂ node 4 · out[0]positive = ◂ node 5 · out[0]negative = ◂ node 6 · out[0]latent_image = ◂ node 7 · out[0]seed = {{seed}} tmplsteps = {{steps}} tmplcfg = 1.0sampler_name = res_multistepscheduler = simpledenoise = 1.0

outputs: LATENT

9VAEDecodecore

samples = ◂ node 8 · out[0]vae = ◂ node 3 · out[0]

outputs: IMAGE

Prompt construction#

template

{base_prompt}. Highly detailed, sharp focus, high quality.

Variables (1)#

base_prompt

{{prompt}} tmpl

Parameter banks (1)#

The prompt / configuration lookup tables this workflow keys into from its inputs — the vocabulary that turns a style / palette / preset selection into graph parameters.

`aspect_map` (3)#

portrait_896x1152

{"width": 896, "height": 1152}

square_1024

{"width": 1024, "height": 1024}

landscape_1152x896

{"width": 1152, "height": 896}

Models & dependencies#

Models required (3)#

z_image_turbo_bf16.safetensors

qwen_3_4b.safetensors

z_image_ae.safetensors

Output contract#

What a successful run of this workflow returns.

type

image

format

png

Taxonomy & routing#

How the control plane classifies this workflow — from the committed workflow-taxonomy-registry.json. It drives the consistency / control surface the agentic director can exercise over the workflow.

assetFamily

portrait-hero-image-bundle

outputPackageProfile

image-single-profile

controlModalities

model-lockpalette-locksampler-scheduler-lockseed-lockstyle-anchoridentity-adapter

consistencyDimensions

identitylightinglens

notes

Flagship 2026 photoreal portrait on Z-Image-Turbo.