This workflow is designed to run the cutting-edge Z-Image Turbo model. It is a next-generation Text-to-Image generator that utilizes the advanced Qwen 2 (Lumina) text encoder for incredibly accurate prompt understanding and Flow Matching architecture for high detail.
The main advantage of this setup is speed. The workflow is configured to generate a 1024x1024 image in just 9 steps, while maintaining photorealistic quality.
🚀 Key Features:
- Qwen/Lumina Text Encoder: Unlike standard SDXL, this uses qwen_3_4b.safetensors. This allows you to write massive, detailed prompts (as seen in the example within the file), and the model will adhere to them with high precision.
- AuraFlow Sampling: The use of the ModelSamplingAuraFlow node adapts the generation process to the specific model architecture, ensuring correct color saturation and contrast.
- Turbo Speed: Thanks to the z_image_turbo model, you only need 9 sampling steps (Scheduler: simple, Sampler: euler), making generation almost instant on powerful GPUs.
- LoRA Support: The workflow already includes a node for loading LoRA (configured for lenovo_z.safetensors with a weight of 0.25 in the example), allowing for easy stylization.