Z-Image
Alibaba Tongyi's 6B Apache 2.0 image model; Turbo draws photoreal images with English and Chinese text in 8 steps on a 16 GB GPU.
In short. Z-Image is an image generator from Alibaba's Tongyi MAI team that is small enough for a gaming graphics card. Its Turbo version makes a picture in just 8 steps and renders English and Chinese text inside images. It has about 6 billion parameters and uses the Apache 2.0 license, so you can use it in products.
Summary
Tongyi MAI, a team in Alibaba Group's Tongyi Lab, released Z-Image-Turbo on 26 November 2025 and the undistilled base model Z-Image on 27 January 2026, both under the Apache 2.0 license. The family is built on a 6B-parameter (6.15B) Scalable Single-Stream Diffusion Transformer, called S3-DiT, which feeds text tokens, visual-semantic tokens and image-latent tokens through one stream. The text encoder is Qwen3-4B, and images are encoded with the Flux VAE. Tongyi says the whole training run took about 314,000 H800 GPU hours, roughly US$630,000.
Turbo is a distilled version that needs only 8 function evaluations and no classifier-free guidance, and Tongyi says it fits in 16 GB of consumer-GPU memory. The base Z-Image uses 50 steps with guidance, supports negative prompts and gives more varied results, so it is the better starting point for fine-tuning. Tongyi presents Turbo as strongest at photorealism and at rendering English and Chinese text. Its GitHub page says Turbo ranked 8th overall, and first among open models, on the Artificial Analysis text-to-image leaderboard on 8 December 2025; that is a snapshot, and rankings move.
This is a text-to-image model. The GitHub page still lists Z-Image-Edit and Z-Image-Omni-Base as "to be released", and no such checkpoints were on Hugging Face when we checked on 11 November 2026. The Turbo repository stores the transformer in FP32, so the whole download is about 33 GB, while the base repository is BF16 at about 20.5 GB (both include the 8 GB text encoder). Hugging Face showed about 743,000 downloads of Turbo in the 30 days to 11 November 2026.
Good for
- Fast local image generation on a single 16 GB consumer GPU
- Photorealistic images and posters with English or Chinese text in them
- Fine-tuning and LoRA training from the base Z-Image checkpoint
- Commercial products, since the Apache 2.0 license has few conditions
Watch out for
- Text-to-image only, because the editing and omni checkpoints are not released yet
- Turbo has lower diversity than the base model, and because it runs without guidance it lacks controls such as negative prompts that the base model offers
- The leaderboard ranking is a December 2025 snapshot reported on Tongyi's own page
- The full download is large (about 33 GB for Turbo), mostly because the transformer is stored in FP32
Performance claims come from the developer and have not been independently verified.
How much memory does it need?
Memory figures for each variant of Z-Image. Image, video, speech and music pipelines also load a text encoder and a decoder, so the memory needed can be larger than the main checkpoint.
| Variants | Parameters | Download size | Memory needed | Fits on |
|---|---|---|---|---|
| Z-Image-Turbo FP32 | 6.15B | 33 GB | 16 GB | Laptop or phone |
| Z-Image (base) BF16 | 6.15B | 21 GB | 24 GB | Gaming GPU |
Download size covers every component needed to run the model, such as the generator, text encoder and decoder. Memory needed is a stated or published measured minimum where one exists; otherwise it is the download size plus about 15%. Larger images, longer clips and higher resolutions need more. The developer states a minimum of 16 GB of GPU memory.
The license in plain words
One of the most widely used permissive licenses. You may use, change and sell products built on the model with only light conditions.
See also
- FLUX.3 Image Klein vs Z-Image: which open image model should you use?
- MOSS-Image vs Z-Image: which Apache 2.0 image model?
- Best open-source image and video generation models
- Krea 2: an open image model trained from scratch for variety
- MOSS-Image: a 5B open image model for long text and JSON layouts
Frequently asked questions
- What is Z-Image?
- Z-Image is an image generator from Alibaba's Tongyi MAI team that is small enough for a gaming graphics card. Its Turbo version makes a picture in just 8 steps and renders English and Chinese text inside images. It has about 6 billion parameters and uses the Apache 2.0 license, so you can use it in products.
- Can I use Z-Image commercially?
- Yes. Z-Image is released under Apache 2.0, a permissive license. You can use it commercially if you follow the standard conditions, such as keeping the license notice.
- Is Z-Image open source?
- Z-Image is released under Apache 2.0, a license approved by the Open Source Initiative. Training data and training code may not be public, so it is best described as open weights under an open-source license.
- Can I run Z-Image on my own computer?
- Yes. The smallest variant, Z-Image-Turbo, needs about 16 GB of memory at 4-bit precision, which fits a modern laptop or a mid-range graphics card.
- When was Z-Image released?
- Z-Image was released by Tongyi MAI on November 26, 2025.
Official links
Sources
Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.
- Z-Image project page (Tongyi MAI) tongyi-mai.github.io
- Z-Image repository, release news (GitHub) github.com
- Z-Image-Turbo model card (Hugging Face) huggingface.co
- Z-Image model card (Hugging Face) huggingface.co
- Z-Image technical report (arXiv 2511.22699) arxiv.org
Release date and facts last verified November 11, 2026.
Related models
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| Qwen-Image-2.1 Alibaba Qwen | 7.1B | – | Qwen Research License (Non-commercial) | 20 GB Gaming GPU | |
| FLUX 3 Image Klein Black Forest Labs | 5.6B | – | FLUX Kommunity (Conditional) | 8 GB Laptop or phone | |
| MiniMax H3-Image MiniMax | 33.1B | – | MiniMax H3-Image Community (Conditional) | 155 GB Big-memory desktop | |
| MOSS-Image OpenMOSS | 5.08B | – | Apache 2.0 (Permissive) | 4.2 GB Laptop or phone |