Alibaba has open-sourced Z-Image, a high-efficiency image generation model with 6 bytes of parameters.

Alibaba has open-sourced Z-Image, a high-efficiency image generation model with 6 bytes of parameters. The model employs a single-stream DiT architecture and improves parameter efficiency by uniformly processing text and visual semantic tokens. Three versions are released: Z-Image-Turbo is a lightweight distillation version that supports 8-step rapid generation and excels at realistic generation and bilingual text rendering; Z-Image-Base is the base model for community fine-tuning and customization; and Z-Image-Edit focuses on image editing, supporting creative editing based on natural language commands.