ComfyUI Workflow for Z Image
Any feedback would be appreciated.
📂 Required Models
Diffusion Model: z_image_turbo_bf16.safetensors .../ComfyUI/models/diffusion_models/
Text Encoder: qwen_3_4b.safetensors .../ComfyUI/models/text_encoders/
VAE: ae.safetensors .../ComfyUI/models/vae/
📂 Optional Models
Upscale Model: 4x_foolhardy_Remacri.pth .../ComfyUI/models/upscale_models
Detection Model: face_yolov9c.pt .../ComfyUI/models/ultralytics/bbox
SAM Model: sam_vit_b_01ec64.pth .../ComfyUI/models/sams
Fun-Controlnet-Union: Z-Image-Turbo-Fun-Controlnet-Union.safetensors .../ComfyUI/models/model_patches
https://huggingface.co/alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union/tree/main
Z Image Fun-Controlnet-Union requires the ComfyUI nightly version as of 3rd Dec. 2025
🧩 Required Custom Nodes
🟩 ComfyUI-Manager (by Comfy-Org)
https://github.com/Comfy-Org/ComfyUI-Manager🟩 rgthree-comfy (by rgthree)
https://github.com/rgthree/rgthree-comfy🟩 ComfyUI-Easy-Use (by yolain)
https://github.com/yolain/ComfyUI-Easy-Use🟩 ComfyUI-Impact-Pack (by ltdrdata)
https://github.com/ltdrdata/ComfyUI-Impact-Pack🟩 ComfyUI-Impact-Subpack (by ltdrdata)
https://github.com/ltdrdata/ComfyUI-Impact-Subpack🟪 ComfyUI-Image-Saver (by alexopus)
https://github.com/alexopus/ComfyUI-Image-Saver🟪 ComfyUI-Lora-Manager (by willmiao)
https://github.com/willmiao/ComfyUI-Lora-Manager🟪 ComfyUI-KJNodes (by kijai)
https://github.com/kijai/ComfyUI-KJNodes
⚙️ Recommended Settings
Turbo Steps: 6 to 10 (Default set to 9).
Turbo CFG: 1.0 (Default set to 1.0).
Base Steps: 25 to 50
Base CFG: 3.0 to 6.0
Sampler: euler or res_multistep
Scheduler: simple
[1:1] 1024x1024
[3:4] 896x1152
[5:8] 832x1216
[9:16] 768x1344
[9:21] 640x1536Experiment and enjoy!
Description
Updated to V5.0
(This is not an exhaustive changelog)
General:
Fixes Improvements
Added MutliPass
FAQ
Comments (8)
The best Z-Image WF just got better, thanks!
nice workfolw
CLIPLoader
Llama2 加载state_dict错误： model.embed_tokens.weight的尺寸不匹配：用形状炬复制参数。从检查点的尺寸（[151936， 2560]）中，当前模型的形状是火炬。尺寸（[128256， 4096]）。 model.layers.0.self_attn.Q_Proj.weight的尺寸不匹配：用形状炬复制参数。从检查点开始，当前型号的形状是火炬。大小（[4096， 4096]）。 model.layers.0.self_attn.k_proj.weight的尺寸不匹配：用形状炬复制参数。从检查点看，尺寸（[1024， 2560]），当前型号的形状是火炬。尺寸（[1024， 4096]）。 model.layers.0.self_attn.v_proj.weight的大小不匹配：用形状炬复制参数。从检查点看，尺寸（[1024， 2560]），当前型号的形状是火炬。尺寸（[1024， 4096]）。 model.layers.0.self_attn.o_proj.weight的尺寸不匹配：用形状炬复制参数。从检查点的尺寸（[2560， 4096]）来看，当前型号的形状是火炬。大小（[4096， 4096]）。
我也是这个问题，兄弟你解决了吗？怎么解决的
Why not a simple Lora Loader. I hate the Manager. Why a manager? The model only allows a very small amount of Loras anyway. That makes no sense at all! Because of this workflow, am I supposed to modify my extra_model_paths.yaml file? Where have all the simple workflows gone...
I have no Idea how this is doing what it is doing but the results are the best I have seen yet from Z image, I did swap out the lora loader for the basic RGTHREE version but the detail and speed is on another level. Many thanks
I can't get it to use a random seed, I mean the output result is great an all, but it is 98% the same image.
That's just Z Image Turbo, but I guess you could use the SeedVarianceEnhancer custom node to trade a bit of prompt adherence for more diversity with the same prompt.