by Alibaba Tongyi-MAI
Ultra-fast image generation model from Alibaba Tongyi-MAI using S3-DiT architecture. 6B parameters, only 8 inference steps. Fits in 16GB VRAM.
VRAM requirements, GPU fit, and setup notes for Z-Image Turbo, including 8GB/12GB fit guidance where relevant. Recommended runtimes: ComfyUI and Diffusers support. Best download size: ~30.5 GB at BF16.
Your hardware
Detecting...
Measured quality metrics for Z-Image Turbo outputs.
How often humans prefer this model's output (0-100%)
Visual quality and composition rating (5-9 scale)
Compare which GPUs can run Z-Image Turbo at different precisions. FP8 uses less memory than FP16 when available, and the grade shows how comfortably each GPU handles the workload.
| Resolution | VRAM Required | RTX 4090 24GB | RTX 3060 12GB | RTX 4060 8GB | MacBook Pro M4 Pro 24GB |
|---|---|---|---|---|---|
| 512×512 | 21.9 GB | A | F | F | D |
| 768×768 | 22.0 GB | A | F | F | D |
| 1024×1024 | 22.3 GB | B | F | F | D |
from diffusers import DiffusionPipeline
import torch
pipe = DiffusionPipeline.from_pretrained(
"Tongyi-MAI/Z-Image-Turbo",
torch_dtype=torch.float16
)
pipe.to("cuda")
image = pipe(
prompt="your prompt here",
num_inference_steps=8,
guidance_scale=7.5,
height=1536,
width=1536,
).images[0]
image.save("output.png")Get started
Setup instructions for running Z-Image Turbo locally
1. Download the model
Get the checkpoint from HuggingFace
2. Place in:
ComfyUI/models/checkpoints/3. Launch ComfyUI
python main.pyVRAM allocation at 1024×1024 on RTX 4090 24GB (24 GB)
Time per image at 1024×1024, 28 steps, FP16.
Download Z-Image Turbo in the precision that matches your GPU. Lower precision usually means less VRAM pressure, while higher precision keeps more quality.
| Format | Precision | Size | Provider | |
|---|---|---|---|---|
| safetensors | BF16 | 30.5 GB | official | Download |
Very new model with limited LoRA support so far.
Frequently asked questions
Z-Image Turbo (6B parameters) requires approximately 22.3 GB of VRAM at FP16 precision for standard 1024×1024 image generation. If you want a lighter setup, lower precisions like FP8 can reduce memory use when available.
Z-Image Turbo usually needs more than 8GB for comfortable local use. Check the VRAM table above for the exact resolution and precision trade-off.
Z-Image Turbo is marked for ComfyUI and Diffusers support in our catalog, so those are the runtimes we recommend first for local setup. If your workflow uses another front end, check the model's available formats and workflow notes above before downloading.
Z-Image Turbo can run on the RTX 4090 with sequential offloading enabled, though generation will be slower than native fit.
There are currently no known ControlNet adapters for Z-Image Turbo. Check Hugging Face and Civitai for community-contributed adapters.
Very new model with limited LoRA support so far. The LoRA ecosystem for Z-Image Turbo is rated as "minimal". Each LoRA adds roughly 0.3 GB of extra VRAM.
On a reference GPU (RTX 4090 24GB), Z-Image Turbo generates a 1024×1024 image in approximately ~4.1s at FP16 with 28 inference steps. Faster GPUs with higher memory bandwidth will produce images more quickly.
See also