Generate Photorealistic Images in Sub-Seconds

Z-Image-Turbo utilizes a Scalable Single-Stream Diffusion Transformer (S3-DiT) architecture and a novel Decoupled-DMD distillation algorithm, enabling photorealistic image generation in just 8 steps (NFEs).

Z-Image-Turbo comfortably runs on consumer devices with 16GB VRAM, making it accessible to a wide range of users without the need for expensive enterprise GPUs.

Compared to models like FLUX.2 and Qwen-Image which require 20B+ parameters, Z-Image-Turbo achieves competitive performance with only 6B parameters, offering a balance of quality and accessibility.

The guidance scale should be set to 0.0 for Z-Image-Turbo models, as they are specifically distilled and optimized for this configuration. Varying from this may produce unpredictable results.