AI that generates 1024×1024 images in under 1 second, with significant improvements in VRAM efficiency

Alibaba’s Tongyi Lab announced the “Z-Image Turbo AI,” which is poised to be a major turning point in image generation technology. The most remarkable feature of this AI model is its overwhelming speed, capable of generating photorealistic images at 1024×1024 pixels in less than 1 second.

Achieving Innovative Lightweight and Efficiency

Z-Image Turbo AI is equipped with 6 billion parameters and compresses inference steps to just 8 steps using Decoupled-DMD distillation technology. This optimization enables fast processing even on consumer GPUs with 16GB VRAM. In the field of image generation, where increasing VRAM was once essential, this model brings a revolution in VRAM efficiency, significantly lowering the access barrier for general users.

Balancing Multilingual Support and Practicality

Licensed under Apache 2.0 and open-sourced, this model features bilingual prompt processing and high-precision rendering of text within images. This greatly expands its potential for use in global markets. Its design, which significantly reduces hardware constraints while improving prompt accuracy, could set a new standard for localized AI deployment.

This technological advancement hints at new pathways for future AI inference optimization and model deployment on edge devices.