Update - Link for Z-Image Supervised FineTune Model in FP32 Link

Z-Image Transformer Setup for ComfyUI

This was built using ComfyUI Github version. Not desktop, portable, or cloud.

Note: The Z-Image transformer is FP32. This method uses a modified script from ComfyUI with QKV mappings. It does not modify supported_models.py. FP32 may be overkill for some setups.

Comfy's script "https://huggingface.co/Comfy-Org/z_image_turbo/blob/main/z_image_convert_original_to_comfy.py"

Prerequisites

• System RAM: Ensure you have at least 32GB of System RAM, as you will be processing a massive 24GB FP32 model.

---

Method 1: Convert and Merge (Recommended)

1. Download Required Files

• Download the script: Z-Image_convert_and_merge.py
• Download the transformer files: Z-Image-Turbo Transformer Files

2. Prepare the Folder

Place the downloaded python script and the transformer files (safe tensors) into the same folder.

3. Run the Merge Command

Open your command line interface (CMD) from that directory and run the following command:

python Z-Image_convert_and_merge.py Z-Image_FP32.safetensors diffusion_pytorch_model-00001-of-00003.safetensors diffusion_pytorch_model-00002-of-00003.safetensors diffusion_pytorch_model-00003-of-00003.safetensors

CRITICAL: Do not include "bf16" or "fp16" in the output filename (Z-Image_FP32.safetensors). If you do, the script will attempt to convert it. We want it to default to FP32.

4. Load in ComfyUI

1. Refresh ComfyUI.
2. Use the "Diffusion Model Loader KJ" custom node.
3. Set the settings to FP32.
4. Load the generated safetensor and enjoy.

---

Method 2: Legacy Method (Modifies Core Files)

This was built using ComfyUI Github version. Not desktop, portable, or cloud.

WARNING: THIS IS A BETA MOD. DO IT AT YOUR OWN RISK. This mod is designed for ComfyUI Version 3.75. It may fail on newer or older versions.

Instructions

1. Backup: Create a backup of your original supported_models.py located in ComfyUI\comfy.
2. Install Mod: Drag and drop the modded version to that folder to support the FP32 Z-Image model.
3. Merge Files: Go to HuggingFace and merge their Transformer files into 1 file. (Requires at least 32GB of CPU RAM to merge).
4. Run: Load your Safetensor and enjoy.

Common Errors

• unet missing: ['norm_final.weight']
  • Note: The image will usually still generate despite this error.