How much GPU VRAM is needed for Z-Image-Turbo image generation?

Running Z-Image-Turbo locally at standard 1024x1024 resolution requires ~12GB to 16GB of VRAM depending on the precision variant (FP8 vs FP16) and text encoder backbone.

Is Z-Image-Turbo open source and free to self-host?

Yes, Z-Image-Turbo provides open weights that can be self-hosted locally on compatible hardware with $0 per-image token fees.

Back to Dashboard

Image ModelOpen Source

Z-Image-Turbo VRAM Requirements

Developed by Wan (Kuaishou / ByteDance)

Determine the local VRAM and system RAM requirements to run the Z-Image-Turbo image generation model. Calculate components footprint, check VRAM margins, and optimize configurations for local generation speed.

Hardware Configuration

Adjust settings to check compatibility with your system in real time.

GPU Dedicated VRAM

GB VRAM

Available: 14.50 GB

System Host RAM

GB RAM

Available: 29.00 GB

CPU Offloading

Compatibility Verdict

Out of Memory Risk

This configuration exceeds your system's usable memory capacity. Attempting to run it will cause crashes or freeze your machine.

Requires 38.79 GB total memory (weights: 32.83 GB, overhead: 5.96 GB).

Memory Margin0 GB

Z-Image-Turbo Precision Variants

Z-Image-Turbo Component VRAM Breakdown

transformer24.61 GB

text encoder8.05 GB

vae0.17 GB

Activation Overhead4.96 GB

Hardware Overhead1.00 GB

Total Active VRAM Required38.79 GB

How to Setup and Run Z-Image-Turbo Locally

Download Weights from Hugging Face

Open-source image model weights are hosted on Hugging Face. You will need to download the model files (safetensors or GGUF variants) and place them in your UI's model folder.

Hugging Face Repository

Setup inside ComfyUI or WebUI

Most image models require a generation framework:

ComfyUI (Recommended): Put the UNet/Transformer weight files inside `ComfyUI/models/checkpoints/` or `ComfyUI/models/unet/` and the VAE inside `ComfyUI/models/vae/`.
Stable Diffusion WebUI (Forge/re-forge): Put checkpoint files inside `webui/models/Stable-diffusion/`.

Model Specs

DeveloperWan (Kuaishou / ByteDance)
AvailabilityLocal-Only
Official SiteVisit site

Key Strengths

realismprompt adherenceeditingoutput qualitycharacter consistencytext rendering

Commercial API Pricing

Commercial API costs $0.0050 per generated image.

Model Release Alerts

Get notified when new models drop that run on your hardware. Zero spam.