Hobbyists and developers wanting an open image model to fine-tune; slower than Turbo, but free under Apache 2.0.
Kenning's take
What it does
Z-Image is an open image generation model with 6 billion parameters from Tongyi-MAI, Alibaba's AI team. It is the non-distilled base of the faster Z-Image-Turbo, trading speed for image quality, variety and control, and is meant for creative work and for fine-tuning. The weights are free to download from Hugging Face and ModelScope under the Apache 2.0 licence, and there is an online demo.
- 6-billion-parameter single-stream diffusion transformer
- Wide range of artistic styles
- Negative prompting support
- High diversity of faces, poses and layouts
- Suited to fine-tuning and LoRA training
- Apache 2.0 open weights
Who it suits. Developers and image-generation hobbyists who want an open model they can run or fine-tune themselves.
Key facts
- Category
- AI models
- Made by
- Alibaba (Tongyi-MAI)
- Based in
- China
- Released
- 27 January 2026
- How to get it
- open weights on Hugging Face and ModelScope; online demo on Hugging Face Spaces
- Works on
- Hugging Face, ModelScope, self-hosted
- In the UK
- Yes
- Free plan
- free to download and run under Apache 2.0
Reviews and write-ups
- GIGAZINE 2026-01-28
FAQ
When was Z-Image released?
The Z-Image base model was released on 27 January 2026, after Z-Image-Turbo in November 2025.
Is Z-Image free?
Yes. The weights are free under the Apache 2.0 licence on Hugging Face and ModelScope.
How is Z-Image different from Z-Image-Turbo?
Z-Image is the full base model, aimed at quality, variety and fine-tuning. Turbo is a distilled version built for speed.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.