Z-Image Controlled Text-to-Image — ComfyUI workflow

Z-Image text-to-image steered by a control image through the Fun-Controlnet-Union 2.1 patch (canny, depth, pose or tile), on the base or a Civitai Turbo finetune, with an optional Z-Image LoRA. Translated from Comfy-Org/workflow_templates@aaac56dd templates/image_z_image_turbo_fun_union_controlnet.json: ModelPatchLoader + QwenImageDiffsynthControlnet (strength 1), core Canny (low 0.1, high 0.32), the control image through ImageScaleToMaxDimension (lanczos, 1024) and GetImageSize into EmptySD3LatentImage, ModelSamplingAuraFlow shift 3, res_multistep + simple, Turbo at 8 steps and CFG 1 with the negative zeroed. The base schedule (25 steps, CFG 4, a real negative) is templates/image_z_image.json's. Depth runs the core Depth Anything 3 nodes with the values of templates/utility_depth_anything3_image_depth_estimation.json. Choosing a Turbo checkpoint selects the Turbo union patch and the 8-step schedule through checkpoint_profile.

25Graph nodes

23Node classes

13Inputs

15Models

Tags: zimagerunpod-serverlessvolume-backedtext-to-imagecontrolnet

Inputs (13)#

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

promptstringrequireddefault Realistic photo, close-up of a latina model peeking through pine branches, dappled sunlight on her face, natural, moody, smooth skin, a little bit film grain.

Positive prompt in natural prose (the default is the upstream union-controlnet template's prompt)

negative_promptstringdefault

Negative prompt, honoured by the base checkpoint (CFG-guided). A Turbo checkpoint samples at CFG 1 with the negative zeroed (ConditioningZeroOut, as the upstream template does), so this text is not encoded for it.

checkpointenumdefault zimage-base-bf16

Z-Image checkpoint on the network volume (see zimage-txt2img). The base loads the Z-Image Fun-Controlnet-Union 2.1 patch; the two Turbo checkpoints load the Turbo Fun-Controlnet-Union 2.1 2602 8-step patch and run 8 steps at CFG 1 with the negative zeroed, whatever steps and cfg say.

options: zimage-base-bf16cyberrealistic-zimage-turbo-v8-bf16zimage-turbo-bf16-aio

stepsintegerdefault 25min 8max 60

Sampling steps for the base checkpoint (25 is the upstream base template's value). Not used by a Turbo checkpoint, which always runs 8.

cfgfloatdefault 4.0min 1.0max 10.0

Classifier-free guidance for the base checkpoint (4 is the upstream base template's value). Not used by a Turbo checkpoint, which always samples at 1.

seedintegerdefault -1

Random seed (-1 = random)

control_imageimagerequired

The image the control is extracted from. It is scaled so its longest side is 1024 (lanczos) and the output takes its aspect ratio at that size, as the upstream template does with ImageScaleToMaxDimension and GetImageSize.

controlenumdefault canny

Which control the union patch receives. canny = ComfyUI's core Canny at the upstream template's thresholds (0.1 / 0.32); depth = core Depth Anything 3 (mono large, 504 px, V2-style near-white depth, the upstream Depth Anything 3 template's values); pose = controlnet_aux DWPose (body, hands and face, the settings the motion catalog's pose control uses); tile = controlnet_aux TilePreprocessor at its node defaults (3 pyrUp iterations, 512 px), since no upstream template sets them.

options: cannydepthposetile

control_strengthfloatdefault 1.0min 0.0max 2.0

Union patch strength (QwenImageDiffsynthControlnet.strength; 1.0 is the upstream template's value). The node accepts -10..10; this catalog offers 0..2.

control_end_fractionfloatdefault 1.0min 0.1max 1.0

The share of the sampling steps that run with the control, ending at step round(steps x fraction) of the effective schedule (25 by default on the base, 8 on Turbo). The union patch node has no start/end input, so below 1.0 the sampler is split into two KSamplerAdvanced passes over the same schedule: the controlled model up to the end step (leftover noise kept), then the same model without the patch to the end, adding no noise (so the seed only acts in the first pass). At 1.0 the second pass is absent and the graph samples once, as the upstream template does. A documented approximation: res_multistep restarts its multistep history at the split.

batch_sizeintegerdefault 1min 1max 1

Images per run. Pinned to 1: the output takes the control image's size at a longest side of 1024 (up to 1 MP), where the executor's resolution-scaled batch cap (C.12.19) allows one latent.

loraenumdefault none

Optional Z-Image LoRA (A.03.02; see zimage-txt2img). none leaves the graph without a LoRA loader.

options: nonephotoreal-betternudes-v3realistic-snapshot-v5breast-sliderdetail-sliderdetaildaemonz-v1.1

lora_strengthfloatdefault 1.0min -2.0max 2.0

LoRA strength on the model (LoraLoaderModelOnly.strength_model).

ComfyUI node graph (25)#

The executable ComfyUI prompt graph: 25 nodes across 23 distinct node classes, wired by 21 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: CheckpointLoaderSimple"] n4["5: LoraLoaderModelOnly"] n5["6: ModelSamplingAuraFlow"] n6["7: CLIPTextEncode"] n7["8: CLIPTextEncode"] n8["9: ConditioningZeroOut"] n9["10: LoadImage"] n10["11: ImageScaleToMaxDimension"] n11["12: GetImageSize"] n12["13: EmptySD3LatentImage"] n13["14: Canny"] n14["15: LoadDA3Model"] n15["16: DA3Inference"] n16["17: DA3Render"] n17["18: DWPreprocessor"] n18["19: TilePreprocessor"] n19["20: ModelPatchLoader"] n20["21: QwenImageDiffsynthControlnet"] n21["22: KSamplerAdvanced"] n22["23: KSamplerAdvanced"] n23["24: VAEDecode"] n24["25: SaveImage"] n6 -->|conditioning| n8 n9 -->|image| n10 n10 -->|image| n11 n11 -->|width| n12 n11 -->|height| n12 n10 -->|image| n13 n14 -->|da3_model| n15 n10 -->|image| n15 n15 -->|da3_geometry| n16 n10 -->|image| n17 n10 -->|image| n18 n5 -->|model| n20 n19 -->|model_patch| n20 n20 -->|model| n21 n6 -->|positive| n21 n12 -->|latent_image| n21 n5 -->|model| n22 n6 -->|positive| n22 n21 -->|latent_image| n22 n22 -->|samples| n23 n23 -->|images| n24

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

Nodes (25)#

1UNETLoadercore

unet_name = {{checkpoint_map[checkpoint]}} tmplweight_dtype = default

outputs: MODEL

2CLIPLoadercore

clip_name = qwen_3_4b.safetensorstype = lumina2device = default

outputs: CLIP

3VAELoadercore

vae_name = ae.safetensors

outputs: VAE

4CheckpointLoaderSimplecore

ckpt_name = {{checkpoint_map[checkpoint]}} tmpl

outputs: MODELCLIPVAE

5LoraLoaderModelOnlycore

model = {{loader_refs[checkpoint_profile[checkpoint].loader].model}} tmpllora_name = {{lora_map[lora]}} tmplstrength_model = {{lora_strength}} tmpl

outputs: MODEL

6ModelSamplingAuraFlowcore

model = {{lora_ref if lora != 'none' else loader_refs[checkpoint_profile[checkpoint].loader].model}} tmplshift = 3

outputs: MODEL

7CLIPTextEncodecore

text = {{constructed_prompt}} tmplclip = {{loader_refs[checkpoint_profile[checkpoint].loader].clip}} tmpl

outputs: CONDITIONING

8CLIPTextEncodecore

text = {{negative_prompt}} tmplclip = {{loader_refs[checkpoint_profile[checkpoint].loader].clip}} tmpl

outputs: CONDITIONING

9ConditioningZeroOutcore

conditioning = ◂ node 7 · out[0]

outputs: CONDITIONING

10LoadImagecore

image = {{control_image}} tmpl

outputs: IMAGEMASK

11ImageScaleToMaxDimensioncore

image = ◂ node 10 · out[0]upscale_method = lanczoslargest_size = 1024

outputs: IMAGE

12GetImageSizecore

image = ◂ node 11 · out[0]

outputs: INTINTINT

13EmptySD3LatentImagecore

width = ◂ node 12 · out[0]height = ◂ node 12 · out[1]batch_size = {{batch_size}} tmpl

outputs: LATENT

14Cannycore

image = ◂ node 11 · out[0]low_threshold = 0.1high_threshold = 0.32

outputs: IMAGE

15LoadDA3Modelcore

model_name = depth_anything_3_mono_large.safetensorsweight_dtype = default

outputs: DA3_MODEL

16DA3Inferencecore

da3_model = ◂ node 15 · out[0]image = ◂ node 11 · out[0]resolution = 504resize_method = upper_bound_resizemode = mono

outputs: DA3_GEOMETRY

17DA3Rendercore

da3_geometry = ◂ node 16 · out[0]output = depthoutput.normalization = v2_styleoutput.apply_sky_clip = false

outputs: IMAGE

18DWPreprocessorcore

image = ◂ node 11 · out[0]detect_hand = enabledetect_body = enabledetect_face = enableresolution = 512bbox_detector = yolox_l.onnxpose_estimator = dw-ll_ucoco_384.onnxscale_stick_for_xinsr_cn = disable

outputs: IMAGEPOSE_KEYPOINT

19TilePreprocessorcore

image = ◂ node 11 · out[0]pyrUp_iters = 3resolution = 512

outputs: IMAGE

20ModelPatchLoadercore

name = {{checkpoint_profile[checkpoint].union_patch}} tmpl

outputs: MODEL_PATCH

21QwenImageDiffsynthControlnetcore

model = ◂ node 6 · out[0]model_patch = ◂ node 20 · out[0]vae = {{loader_refs[checkpoint_profile[checkpoint].loader].vae}} tmplimage = {{control_source_map[control]}} tmplstrength = {{control_strength}} tmpl

outputs: MODEL

22KSamplerAdvancedcore

model = ◂ node 21 · out[0]add_noise = enablenoise_seed = {{seed}} tmplsteps = {{distilled_schedule.steps if checkpoint_profile[checkpoint].distilled else steps}} tmplcfg = {{distilled_schedule.cfg if checkpoint_profile[checkpoint].distilled else cfg}} tmplsampler_name = res_multistepscheduler = simplepositive = ◂ node 7 · out[0]negative = {{checkpoint_profile[checkpoint].negative}} tmpllatent_image = ◂ node 13 · out[0]start_at_step = 0end_at_step = {{int(round((distilled_schedule.steps if checkpoint_profile[checkpoint].distilled else steps) * control_end_fraction))}} tmplreturn_with_leftover_noise = {{'enable' if control_end_fraction < 1.0 else 'disable'}} tmpl

outputs: LATENT

23KSamplerAdvancedcore

model = ◂ node 6 · out[0]add_noise = disablenoise_seed = {{seed}} tmplsteps = {{distilled_schedule.steps if checkpoint_profile[checkpoint].distilled else steps}} tmplcfg = {{distilled_schedule.cfg if checkpoint_profile[checkpoint].distilled else cfg}} tmplsampler_name = res_multistepscheduler = simplepositive = ◂ node 7 · out[0]negative = {{checkpoint_profile[checkpoint].negative}} tmpllatent_image = ◂ node 22 · out[0]start_at_step = {{int(round((distilled_schedule.steps if checkpoint_profile[checkpoint].distilled else steps) * control_end_fraction))}} tmplend_at_step = 10000return_with_leftover_noise = disable

outputs: LATENT

24VAEDecodecore

samples = ◂ node 23 · out[0]vae = {{loader_refs[checkpoint_profile[checkpoint].loader].vae}} tmpl

outputs: IMAGE

25SaveImagecore

images = ◂ node 24 · out[0]filename_prefix = isis/zimage-txt2img-control

Prompt construction#

template

{base_prompt}

Variables (1)#

base_prompt

{{prompt}} tmpl

Parameter banks (9)#

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

`option_policy` (2)#

checkpoint

{"zimage-base-bf16": {"distribution": "service", "content": "neutral"}, "cyberrealistic-zimage-turbo-v8-bf16": {"distribution": "service", "content": "racy"}, "zimage-turbo-bf16-aio": {"distribution": "service", "content": "neutral"}}

lora

{"photoreal-betternudes-v3": {"distribution": "service", "content": "explicit"}, "realistic-snapshot-v5": {"distribution": "service", "content": "neutral"}, "breast-slider": {"distribution": "service", "content": "racy"}, "detail-slider": {"distribution": "service", "content": "neutral"}, "detaildaemonz-v1.1": {"distribution": "service", "content": "neutral"}}

`checkpoint_map` (3)#

zimage-base-bf16

z_image_bf16.safetensors

cyberrealistic-zimage-turbo-v8-bf16

cyberrealistic_zimage_turbo_v8_bf16.safetensors

zimage-turbo-bf16-aio

zimage_turbo_bf16_aio.safetensors

`checkpoint_profile` (3)#

zimage-base-bf16

{"loader": "unet", "distilled": false, "negative": ["8", 0], "union_patch": "Z-Image-Fun-Controlnet-Union-2.1.safetensors"}

cyberrealistic-zimage-turbo-v8-bf16

{"loader": "unet", "distilled": true, "negative": ["9", 0], "union_patch": "Z-Image-Turbo-Fun-Controlnet-Union-2.1-2602-8steps.safetensors"}

zimage-turbo-bf16-aio

{"loader": "aio", "distilled": true, "negative": ["9", 0], "union_patch": "Z-Image-Turbo-Fun-Controlnet-Union-2.1-2602-8steps.safetensors"}

`distilled_schedule` (2)#

steps

8

cfg

1.0

`loader_refs` (2)#

unet

{"model": ["1", 0], "clip": ["2", 0], "vae": ["3", 0]}

aio

{"model": ["4", 0], "clip": ["4", 1], "vae": ["4", 2]}

`lora_ref` (2)#

50

`control_source_map` (4)#

canny

140

depth

170

pose

180

tile

190

`lora_map` (6)#

none

photoreal-betternudes-v3

zimage-photoreal-betternudes-v3.safetensors

realistic-snapshot-v5

zimage-realistic-snapshot-v5.safetensors

breast-slider

zimage-breast-slider.safetensors

detail-slider

zimage-detail-slider.safetensors

detaildaemonz-v1.1

zimage-detaildaemonz-v1.1.safetensors

`requires_families` (5)#

zimagezimage-cyberrealisticzimage-aiozimage-controlcontrolnet-aux

Models & dependencies#

Custom node packs (1)#

The non-core ComfyUI node classes this graph requires; the RunPod worker image the workflow runs on must bake or install a pack that provides every one of them.

comfyui_controlnet_aux

Models required (15)#

z_image_bf16.safetensors

cyberrealistic_zimage_turbo_v8_bf16.safetensors

zimage_turbo_bf16_aio.safetensors

qwen_3_4b.safetensors

ae.safetensors

Z-Image-Fun-Controlnet-Union-2.1.safetensors

Z-Image-Turbo-Fun-Controlnet-Union-2.1-2602-8steps.safetensors

depth_anything_3_mono_large.safetensors

yolox_l.onnx

dw-ll_ucoco_384.onnx

zimage-photoreal-betternudes-v3.safetensors

zimage-realistic-snapshot-v5.safetensors

zimage-breast-slider.safetensors

zimage-detail-slider.safetensors

zimage-detaildaemonz-v1.1.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

controlnet-cannycontrolnet-depthcontrolnet-posemodel-locksampler-scheduler-lockseed-lock

consistencyDimensions

identitylightingenvironment

notes

Z-Image with a Fun-Controlnet-Union 2.1 control image (canny, depth, pose, tile); a Turbo checkpoint selects the Turbo patch and the 8-step schedule.

Inputs (13)#

ComfyUI node graph (25)#

Nodes (25)#

Prompt construction#

Variables (1)#

Parameter banks (9)#

option_policy (2)#

checkpoint_map (3)#

checkpoint_profile (3)#

distilled_schedule (2)#

loader_refs (2)#

lora_ref (2)#

control_source_map (4)#

lora_map (6)#

requires_families (5)#

Models & dependencies#

Custom node packs (1)#

Models required (15)#

Output contract#

Taxonomy & routing#

`option_policy` (2)#

`checkpoint_map` (3)#

`checkpoint_profile` (3)#

`distilled_schedule` (2)#

`loader_refs` (2)#

`lora_ref` (2)#

`control_source_map` (4)#

`lora_map` (6)#

`requires_families` (5)#