Z-Image-Turbo Local GUI (Windows) ⚡

A lightning-fast local Gradio Web-UI for the Z-Image-Turbo model. Generate photorealistic images in milliseconds using the power of your local GPU. With support for custom LoRAs.

Optimized for NVIDIA RTX 50-Series (Blackwell) & CUDA 13.0.

(Generated locally in seconds)

---

✨ Features

• ⚡ Turbo Speed: Generates high-quality images in just 4-8 steps.
• 📦 Fully Portable: Embedded Python — no system Python installation required.
• 📊 Real-Time Monitor: Live display of CPU, RAM, and VRAM usage.
• 🖼️ Full Control: Adjustable Resolution (up to 2048x2048), Steps, and Seed.
• 💾 Auto-Save: Results are automatically saved to the outputs folder.
• 📂 Quick Access: Open the output folder directly from the UI.
• 🛄 Local Model Cache: Model is downloaded once to model_cache/ — no internet needed after first run.
• 🎨 LoRA Support: Load custom LoRA styles on-the-fly! Drop any .safetensors LoRA into the loras/ folder and activate it with a single click — no restart required. Fine-tune your output with adjustable LoRA weights from 0.0 to 2.0.

---

📋 Prerequisites

• OS: Windows 10/11
• GPU: NVIDIA RTX 3090 / 4090 / 5090 (16GB+ VRAM recommended)

---

⚙️ Installation

1. Download this repository as a ZIP file and extract it.
2. Double-click install.bat.
   • The script automatically downloads an isolated Python 3.11 environment.
   • It installs PyTorch Nightly (required for Blackwell / RTX 50 Series support).
3. Wait until the installation is complete.

No system Python and no virtual environment required.

---

🚀 Usage

1. Double-click start_z_image_lokal_gui.bat.
2. On the first run, the model (~30GB) is downloaded directly into the model_cache/ folder inside the project directory.
3. On every subsequent run, the model loads instantly from the local cache — no internet connection required.
4. The GUI will open automatically in your browser (usually http://127.0.0.1:7860).

---

🎨 LoRA — Custom Styles & Fine-Tuning

LoRAs let you push your image generation into completely new directions — unique art styles, specific characters, hyper-detailed textures, or cinematic looks. All without retraining the base model.

How to use:

1. Download any compatible .safetensors LoRA file (e.g. from Civitai or Hugging Face).
2. Drop it into the loras/ folder.
3. Open the 🎨 LoRA (Optional) panel in the GUI.
4. Click 🔄 to refresh the list, then select your LoRA.
5. Adjust the LoRA Weight slider to control the effect strength (default: 1.0).

Switch between LoRAs or turn them off at any time — no restart needed.

Same prompt — with and without LoRA:

Without LoRA	With LoRA (Anime Style)

---

⚙️ Recommended Settings

• Steps: 4-8 steps is the sweet spot.
• Resolution: 1024×1024 works best.

---

📂 Folder Structure

After installation and first run, your folder will look like this:

Z-Image-Local-GUI/
├── app.py                          # Main application
├── install.bat                     # One-click installer
├── start_z_image_lokal_gui.bat     # Launch script
├── requirements.txt                # Python dependencies
├── python_env/                     # Isolated Python 3.11 (created by install.bat)
├── model_cache/                    # Z-Image-Turbo model (~30GB, downloaded on first run)
├── loras/                          # Drop your .safetensors LoRA files here
└── outputs/                        # Generated images (auto-created)

---

🔧 Troubleshooting

• OOM (Out of Memory) – Ensure you don't have other heavy GPU apps running.
• Python environment not found – Make sure you ran install.bat before starting the app.

---

🔗 Credits

• Model: Tongyi-MAI/Z-Image-Turbo
• Library: Hugging Face Diffusers

---

🤝 Support

This is a free open-source project. I don't ask for donations. However, if you want to say "Thanks", check out my profile on Spotify. A follow or a listen is the best way to support me! 🎧