Skip to content
latestai.eulatestai.eu
ZI

Z-Image

Try itWebsiteDocsGitHubHugging FacePaper

Open-weights 6B text-to-image model family from Alibaba's Tongyi-MAI lab, with a fast 8-step Turbo variant.

Available in EUImage GenerationModel

Updated EU status checked Last shown

Z-Image at a glance

What is Z-Image?
Z-Image is a model. Open-weights 6B text-to-image model family from Alibaba's Tongyi-MAI lab, with a fast 8-step Turbo variant.
Does Z-Image work in the EU?
Available in the EU. Weights are publicly downloadable from Hugging Face and GitHub under Apache-2.0, which has no geographic restrictions, so they can be run locally in the EU. No EU-specific exclusion was found. Last verified 14 Nov 2026.
How much does Z-Image cost?
Open source, open source. Free to download and use under Apache-2.0; free online demos on Hugging Face Spaces and ModelScope. Z-Image-Turbo was released 2025-11-26; the base Z-Image checkpoint on 2026-01-27. Licence: Apache-2.0.
Where was Z-Image shown?
In 2 creator videos by AI Search. Most recently by AI Search on 20 Oct 2026, at 2:30 in “New top local AI image generator is here!”.

Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.

About

Z-Image is a 6B-parameter image generation model family built on a single-stream diffusion transformer, published by the Tongyi-MAI team (Alibaba Tongyi Lab). The README lists four variants: Z-Image-Turbo (distilled, 8 NFEs), Z-Image (undistilled foundation model), Z-Image-Omni-Base (generation and editing) and Z-Image-Edit (instruction-based image editing). At the time of the provided material, Omni-Base and Edit were marked "to be released".

Z-Image-Turbo targets speed: sub-second latency on H800 GPUs and operation within 16 GB VRAM on consumer hardware, with photorealism and bilingual English/Chinese text rendering. The base Z-Image model uses full classifier-free guidance with 28–50 steps, supports negative prompting, produces more diverse outputs, and is positioned as a starting point for LoRA training, ControlNet and other fine-tuning. Turbo is not fine-tunable and has no negative prompting. Recommended resolution is 512×512 to 2048×2048.

Weights are downloadable from Hugging Face and ModelScope under Apache-2.0 and run through the diffusers ZImagePipeline. Online demos are hosted on Hugging Face Spaces and ModelScope. The Turbo model was reported to rank 8th overall on the Artificial Analysis text-to-image leaderboard (first among open-source models) in December 2025. A technical report is on arXiv (2511.22699). One YouTube reviewer described it as a close competitor to a newer local model, which he judged better in understanding and flexibility.

Key features
  • 6B-parameter single-stream diffusion transformer
  • Z-Image-Turbo: 8 steps, fits in 16 GB VRAM
  • Bilingual English and Chinese text rendering
  • Base model with CFG and negative prompting, 28–50 steps
  • Suited to LoRA and ControlNet fine-tuning
  • Apache-2.0 open weights
2 moments

As seen on YouTube

Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.

Keep exploring

Similar AI tools

Closest neighbours by meaning, not just by category.

LL

LLaDA-Image

Model · inclusionAI

1

Open-weights 6B unified text-to-image and instruction-based image editing model family from inclusionAI.

Available in EUImage Generation
4 Sept 2026Open source
QI

Qwen Image 2.1 Turbo

Model · Alibaba

1

Open-weight accelerated variant of Alibaba's Qwen Image 2.1 for faster text-to-image generation and image editing.

Available in EUImage Generation
Release date n/aOpen source
I1

i1

Model · Princeton

1

Fully open 3B text-to-image diffusion model from Princeton with released weights, training code, and data pipeline.

Available in EUImage Generation
Release date n/aOpen source
MO

MOSS-Image

Model · OpenMOSS

3

Open-weights 5B text-to-image model family from OpenMOSS with native JSON bounding-box layout control and text rendering.

Available in EUImage Generation
16 Oct 2026Open source
MH

MiniMax H3 Image

Model · MiniMax

3

Open-weights MiniMax image generation and editing model producing up to ~16 MP images in 6 sampling steps.

Available in EUImage Generation
3 Nov 2026Open source
BI

Boogu Image

Model

1

Open image generation and editing model family (base, turbo, edit) released under Apache 2.0.

Available in EUImage Generation
Release date n/aOpen source