import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("Tongyi-MAI/Z-Image-Turbo,ostris/Z-Image-De-Turbo", dtype=torch.bfloat16, device_map="cuda")
pipe.load_lora_weights("tutututututu/Z-Image-De-Turbo")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]Z-Image-De-Turbo
📦 This repo is mirrored from ostris/Z-Image-De-Turbo, with complete Diffusers format files added.
This is a pre-de-distilled version of Tongyi-MAI/Z-Image-Turbo, optimized for LoRA training.
📌 Why This Model?
Z-Image-Turbo is a distilled model. Training LoRA directly on it results in blurry, low-quality outputs.
Two solutions exist:
- Use Training Adapter: Load zimage_turbo_training_adapter during training
- Use Pre-de-distilled Model (this repo): The training adapter is pre-merged, ready for direct training
✅ Advantages
- Ready to use: No need to load additional training adapters
- Diffusers format: Complete directory structure, compatible with various training tools
- Inference: Recommended 25 steps, guidance_scale 3.0
📝 Credits
- Original pre-de-distilled model: ostris/Z-Image-De-Turbo
- Base model: Tongyi-MAI/Z-Image-Turbo
- Training adapter: ostris/zimage_turbo_training_adapter
- Downloads last month
- 1
Model tree for tutututututu/Z-Image-De-Turbo
Base model
Tongyi-MAI/Z-Image-Turbo