Extensions/ggf-ltp-zimage
ComfyUI Extension

ggf-ltp-zimage

ComfyUI custom nodes for running L2P Z-Image 6B pixel-space generation, wrapping the public pipeline to enable direct no-VAE model inference.

By gjnave·Created 3 months ago·Updated 3 months ago· 4
gjnave/ggf-ltp-zimage
Nodes—
On cloudLocal install
Stars4
Updated3 months ago
Readme

ggf-ltp-zimage

ComfyUI custom nodes and starter workflows for running L2P Z-Image 6B pixel-space generation in ComfyUI.

Built for people who want to get moving fast: getgoingfast.pro

Example L2P Z-Image output

What This Is

This adds two ComfyUI nodes:

  • L2P Z-Image Pipeline Loader
  • L2P Z-Image Generate

It wraps the public L2P Z-Image pipeline so ComfyUI can run the no-VAE pixel-space 6B model directly and output a normal IMAGE tensor for preview/save nodes.

Included

  • __init__.py custom node entrypoint
  • diffsynth/ runtime files adapted from the public L2P Z-Image HF Space
  • requirements.txt
  • workflows/ggf_l2p_zimage_6b_no_vae.json

Install

From your ComfyUI custom_nodes directory:

cd <ComfyUI>\custom_nodes
git clone https://github.com/gjnave/ggf-ltp-zimage.git
cd ggf-ltp-zimage
<ComfyUI>\venv\Scripts\python.exe -m pip install -r requirements.txt

Restart ComfyUI after installing.

Required Models

This repo does not include model weights. Put these required files in your ComfyUI models folder:

ComfyUI/
  models/
    diffusion_models/
      Z-image-6b-no-VAE.safetensors
    text_encoders/
      qwen_3_4b.safetensors
      Z-Image-Turbo-tokenizer/
        tokenizer/
          merges.txt
          tokenizer.json
          tokenizer_config.json
          vocab.json

The default workflow expects these exact local names:

Z-image-6b-no-VAE.safetensors
qwen_3_4b.safetensors
Z-Image-Turbo-tokenizer/tokenizer

If your file or folder names differ, select the correct entries in the L2P Z-Image Pipeline Loader node.

Download Required Models

Set COMFY_ROOT to your ComfyUI folder, then run the commands below in PowerShell. Replace the placeholder path with the folder that contains your main.py.

$COMFY_ROOT = "<path-to-your-ComfyUI>"
$HF = "$COMFY_ROOT\venv\Scripts\hf.exe"

Install this custom node first, because its requirements.txt installs the Hugging Face hf download CLI:

cd $COMFY_ROOT\custom_nodes
git clone https://github.com/gjnave/ggf-ltp-zimage.git
cd ggf-ltp-zimage
$COMFY_ROOT\venv\Scripts\python.exe -m pip install -r requirements.txt

Download the L2P Z-Image 6B no-VAE model:

Manual download: download model-1k-merge.safetensors, put it in ComfyUI/models/diffusion_models, and rename it to Z-image-6b-no-VAE.safetensors.

& $HF download zhen-nan/L2P model-1k-merge.safetensors --local-dir "$COMFY_ROOT\models\diffusion_models"
Move-Item -Force "$COMFY_ROOT\models\diffusion_models\model-1k-merge.safetensors" "$COMFY_ROOT\models\diffusion_models\Z-image-6b-no-VAE.safetensors"

Download the Qwen 3 4B text encoder:

Manual download: download qwen_3_4b.safetensors and put it directly in ComfyUI/models/text_encoders.

& $HF download Comfy-Org/z_image_turbo split_files/text_encoders/qwen_3_4b.safetensors --local-dir "$COMFY_ROOT\models\text_encoders"
Move-Item -Force "$COMFY_ROOT\models\text_encoders\split_files\text_encoders\qwen_3_4b.safetensors" "$COMFY_ROOT\models\text_encoders\qwen_3_4b.safetensors"
Remove-Item -Recurse -Force "$COMFY_ROOT\models\text_encoders\split_files"

Download the Z-Image tokenizer:

Manual download: download all four tokenizer files and put them in ComfyUI/models/text_encoders/Z-Image-Turbo-tokenizer/tokenizer.

& $HF download Tongyi-MAI/Z-Image-Turbo tokenizer/merges.txt tokenizer/tokenizer.json tokenizer/tokenizer_config.json tokenizer/vocab.json --local-dir "$COMFY_ROOT\models\text_encoders\Z-Image-Turbo-tokenizer"

After downloading, restart ComfyUI. The loader dropdowns should show:

model_name: Z-image-6b-no-VAE.safetensors
text_encoder_name: qwen_3_4b.safetensors
tokenizer_name: Z-Image-Turbo-tokenizer/tokenizer

Workflow

Copy or open:

workflows/ggf_l2p_zimage_6b_no_vae.json

The tokenizer_name dropdown is populated by scanning:

ComfyUI/models/text_encoders

It shows tokenizer folders as relative names, for example:

Z-Image-Turbo-tokenizer/tokenizer

It should not show machine-specific absolute drive paths.

On the original development machine this generated successfully through the ComfyUI API.

Credits

  • Tencent Youtu Research L2P project
  • zhen-nan L2P model release
  • multimodalart L2P Z-Image reference Space
  • Packaged for fast ComfyUI use by GetGoingFast

License

Apache-2.0. See LICENSE and NOTICE.