Z-Image Turbo

Z-Image Turbo is Alibaba's breakthrough text-to-image model. Using a Single-Stream DiT (S3-DiT) architecture and Decoupled-DMD distillation, it generates 1024×1024 images in just 8 inference steps - sub-second on H800 GPUs. Apache 2.0 licensed with 1.06M monthly downloads.

Why Z-Image Turbo stands out

1. 8 inference steps - sub-second generation
2. 6B parameters - lightweight DiT
3. 16GB VRAM - fits consumer GPUs
4. Apache 2.0 - fully open for commercial use
5. Bilingual text - excellent Chinese and English rendering
6. 701 adapters - growing LoRA ecosystem
7. 1.06M monthly downloads - rapidly growing

Architecture

Usage

from diffusers import ZImagePipeline
import torch
 
pipe = ZImagePipeline.from_pretrained(
    "Tongyi-MAI/Z-Image-Turbo",
    torch_dtype=torch.bfloat16,
).to("cuda")
 
image = pipe(
    "A photorealistic cat on a windowsill, sunset lighting",
    height=1024, width=1024,
    num_inference_steps=8,
    guidance_scale=0.0,
).images[0]
image.save("output.png")

When to use

• Fast iteration - instant image generation
• Commercial use - Apache 2.0 license
• Bilingual text - English and Chinese rendering
• Photorealism - excels at realistic images