alibaba-pai/Z-Image-Fun-Lora-Distill · Hugging Face

Model Card

Model Features

• This is a Distill LoRA for Z-Image that distills each steps and CFG. It doesn’t use any Z-Image-Turbo associated weights and is educated from scratch. It is appropriate with different Z-Image LoRAs and Controls.
• This mannequin will barely scale back the output high quality and alter the output composition of the mannequin. For particular comparisons, please discuss with the Results part. In most circumstances, the Distill LoRA performs effectively; at present, the most important subject is that it could make the generated outcomes brighter.
• The objective of this mannequin is to supply quick era compatibility for Z-Image by-product fashions, to not change Z-Image-Turbo.

TODO

• Optimize the output brightness;
• Train a 4-step LoRA.

Results

Work itself

Output 25 steps	Output 8 steps

Output 25 steps	Output 8 steps

Output 25 steps	Output 8 steps

Output 25 steps	Output 8 steps

Work with Controlnet

Pose + Inpaint	Output 25 steps	Output 8 steps

Pose + Inpaint	Output 25 steps	Output 8 steps

Pose	Output 25 steps	Output 8 steps

Pose	Output 25 steps	Output 8 steps

Canny	Output	Output 8 steps

Depth	Output	Output 8 steps

Inference

Go to the VideoX-Fun repository for extra particulars.

Please clone the VideoX-Fun repository and create the required directories:

# Clone the code
git clone 

# Enter VideoX-Fun's listing
cd VideoX-Fun

# Create mannequin directories
mkdir -p fashions/Diffusion_Transformer
mkdir -p fashions/Personalized_Model

Then obtain the weights into fashions/Diffusion_Transformer and fashions/Personalized_Model.

📦 fashions/
├── 📂 Diffusion_Transformer/
│   └── 📂 Z-Image/
├── 📂 Personalized_Model/
│   ├── 📦 Z-Image-Fun-Lora-Distill-8-Steps.safetensors
│   ├── 📦 Z-Image-Fun-Controlnet-Union-2.1.safetensors
│   └── 📦 Z-Image-Fun-Controlnet-Union-2.1-lite.safetensors

Set the lora_path=”Personalized_Model/Z-Image-Fun-Lora-Distill-8-Steps.safetensors” in examples/z_image_fun/predict_t2i_control_2.1.py and examples/z_image_fun/predict_i2i_inpaint_2.1.py

Then run the file examples/z_image_fun/predict_t2i_control_2.1.py and examples/z_image_fun/predict_i2i_inpaint_2.1.py.