v2.0 Turbo Model is Live

Dream it. Type it.Visualize in Seconds.

Unleash the power of Z-Image Turbo. The next-generation distillation model that delivers high-fidelity imagery with 50% fewer steps and 2x the speed.

< 1.5sInference Time
2M+Daily Generations
3.2GBModel Size
4.9/5Community Rating

Image Studio

Professional-grade controls for precise generation.

STATUS:CONNECTED
90 chars
8
3

Ready to Create

Enter your prompt on the left sidebar and adjust the settings to generate unique visuals.

Made with Turbo

Explore what's possible with the Z-Image engine.

Futuristic concept art #102

Steps: 8S: 45221

Futuristic concept art #103

Steps: 8S: 45221

Futuristic concept art #104

Steps: 8S: 45221

Futuristic concept art #105

Steps: 8S: 45221
Architecture

Under The Hood

Built on top of the latest Latent Consistency Models, Z-Image Turbo redefines the balance between computational cost and artistic fidelity.

Lightning Fast Inference

Optimized for speed using Turbo diffusion steps. Generate full HD images in under 2 seconds on standard GPUs.

High Fidelity Details

Despite the low step count, our proprietary VAE decoder retains intricate textures and lighting information.

Precise Control

Full control over seed, flow-shift parameters, and custom aspect ratios gives you the exact composition you need.

Architecture
DiT-XL/2
Parameters
3.5 Billion
Context Window
77 Tokens
License
Apache 2.0

Frequently Asked Questions

Q. What is Z-Image Turbo?

It is a distilled text-to-image model optimized for fewer inference steps (4-8 steps) while maintaining high quality comparable to larger models.

Q. Is it free to use?

Yes, this demo interface connects to the public Gradio space provided by Tongyi-MAI for research and testing purposes.

Q. What are the optimal settings?

For photorealism, we recommend 8 steps with a Shift value of 3.0. For artistic styles, 4-6 steps with Shift 1.0 often yields more creative results.

Q. Can I use the images commercially?

Images generated are subject to the license of the underlying model. Generally, they are open for use, but please check the specific license on HuggingFace.