Back to Gap Map Model components / Image, video, 3D & music generation

Z-Image

open weights / Overall score: 3.0

Z-Image is a 6B-parameter single-stream diffusion transformer for text-to-image generation from Alibaba's Tongyi-MAI lab. Z-Image-Turbo is a distilled version that generates in eight steps and fits in 16 GB of GPU memory, with photorealistic output and bilingual English and Chinese text rendering; the undistilled Z-Image base model supports full classifier-free guidance for fine-tuning.

Openness

3 high confidence

3.0

weights: open(Z-Image-Turbo and Z-Image download from Hugging Face without a gate)
data: closed(the cards and README neither release nor describe the training data)
code: partial(the repository holds inference code only
license: Apache-2.0(OSI

Both Z-Image checkpoints are released under Apache 2.0, with no gate on either. The first-party repository holds inference code only, and the training data is not published.

https://cdn.jsdelivr.net/gh/Tongyi-MAI/Z-Image@main/LICENSE recorded 2026-09-26
Repository LICENSE body: "Apache License Version 2.0, January 2004".
https://cdn.jsdelivr.net/gh/Tongyi-MAI/Z-Image@main/README.md recorded 2026-09-26
README: "DiffSynth-Studio has provided more support for Z-Image, including LoRA training, full training, distillation training, and low-VRAM inference"; no training dataset is named or released.
https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?expand[]=downloads&expand[]=likes&expand[]=cardData&expand[]=gated&expand[]=lastModified&expand[]=createdAt recorded 2026-09-26
Hugging Face API record: cardData license apache-2.0, gated false.
https://huggingface.co/api/models/Tongyi-MAI/Z-Image?expand[]=downloads&expand[]=likes&expand[]=cardData&expand[]=gated&expand[]=lastModified&expand[]=createdAt recorded 2026-09-26
Hugging Face API record: cardData license apache-2.0, gated false.
https://ungh.cc/repos/Tongyi-MAI/Z-Image/files/main recorded 2026-09-26
Repository file list: inference.py, batch_inference.py and the model sources under src/zimage; no training scripts.

Adoption

3 high confidence

3.0

Adoption is measured as Hugging Face downloads of the two checkpoints, nearly all of them for Z-Image-Turbo.

https://huggingface.co/api/models/Tongyi-MAI/Z-Image-Turbo?expand[]=downloads&expand[]=likes&expand[]=cardData&expand[]=gated&expand[]=lastModified&expand[]=createdAt recorded 2026-09-26
Hugging Face API record for Tongyi-MAI/Z-Image-Turbo: downloads 609489.
https://huggingface.co/api/models/Tongyi-MAI/Z-Image?expand[]=downloads&expand[]=likes&expand[]=cardData&expand[]=gated&expand[]=lastModified&expand[]=createdAt recorded 2026-09-26
Hugging Face API record for Tongyi-MAI/Z-Image: downloads 59864.

Capability

3 medium confidence

3.0

Z-Image Turbo places in the upper half of the text-to-image arena, a strong result for a model small enough for a consumer GPU. It trails FLUX.2 [dev] there and sits in the same group of open models.

https://artificialanalysis.ai/image/leaderboard/text-to-image recorded 2026-09-26
Arena leaderboard entries "Z-Image Turbo" (rank 76, Elo 940.4, open weights at huggingface.co/Tongyi-MAI/Z-Image-Turbo) and "Z-Image Base" (rank 111).

Verified 2026-09-26