--- source: 'https://howaiworks.ai/models/z-image' section: models title: Z-Image description: >- Alibaba Tongyi-MAI's 6B open-source image model. Z-Image-Turbo, released November 26, 2025, generates 8-step images inside 16 GB of VRAM under Apache 2.0. tags: - Z-Image - Alibaba - Tongyi-MAI - Image Generation - Text-to-Image - Diffusion Model - Open Source - Apache 2.0 - Chinese AI - Latest category: Image Generation Models developer: Alibaba Tongyi-MAI developerWebsite: 'https://github.com/Tongyi-MAI' modelType: Image Generation Model releaseDate: '2025-11-26' lastUpdated: '2026-07-08' license: Apache 2.0 officialWebsite: 'https://tongyi-mai.github.io/Z-Image-blog/' docsPage: 'https://github.com/Tongyi-MAI/Z-Image' --- # Z-Image > Alibaba Tongyi-MAI's 6B open-source image model. Z-Image-Turbo, released November 26, 2025, generates 8-step images inside 16 GB of VRAM under Apache 2.0. ## Overview **Z-Image** is Alibaba Tongyi-MAI's open-source [image generation](https://howaiworks.ai/glossary/image-generation) model family. The first checkpoint, **Z-Image-Turbo**, was **released on November 26, 2025**, per the project's own News log; the undistilled base model **Z-Image** followed on **January 27, 2026**. The technical report opens by naming the problem it exists to solve, and it is worth quoting exactly: > "The landscape of high-performance image generation models is currently dominated by proprietary systems, such as Nano Banana Pro and Seedream 4.0. Leading open-source alternatives, including Qwen-Image, Hunyuan-Image-3.0 and FLUX.2, are characterized by massive parameter counts (20B to 80B), making them impractical for inference, and fine-tuning on consumer-grade hardware." Z-Image's answer is an **efficient 6B-parameter foundation generative model built upon a Scalable Single-Stream Diffusion Transformer (S3-DiT) architecture that challenges the "scale-at-all-costs" paradigm.** The authors report completing "the full training workflow in just 314K H800 GPU hours (approx. $630K)" — an unusually specific and unusually low training-cost disclosure. The distilled Turbo variant offers "both sub-second inference latency on an enterprise-grade H800 GPU and compatibility with consumer-grade hardware (<16GB VRAM)." That accessibility story is the whole model. Z-Image-Turbo has **933,580 Hugging Face downloads in the trailing 30 days** (6,660,069 all-time) and **4,944 likes**, making it the third most-downloaded `text-to-image` model on the Hub. It is also, verifiably, Apache 2.0 — not "open weights, restricted license." What it is *not* is the best image model available, and the [Performance](#performance-benchmarks) section below says so with numbers. ### A note on naming Three clarifications, because the family's own documentation invites confusion. **"Z-Image" names both the family and one specific checkpoint.** `Tongyi-MAI/Z-Image` is the undistilled base model; `Tongyi-MAI/Z-Image-Turbo` is the 8-step distillation. They are different weights with different behaviour. arena.ai's `modelKey` uses the bare string `z-image` when it means Turbo — its row reads `"modelKey":"z-image","modelDisplayName":"z-image-turbo"`. **Z-Image-Edit and Z-Image-Omni-Base do not exist as downloadable weights.** Both are described in the model card, illustrated with showcase images, and Z-Image-Edit is named in the technical report's abstract ("our omni-pre-training paradigm also enables efficient training of Z-Image-Edit"). But the Model Zoo table has marked both **"To be released"** since launch, the GitHub repository has not been pushed to since **February 9, 2026**, and the Tongyi-MAI Hugging Face organisation holds exactly four repositories: `Z-Image-Turbo`, `Z-Image`, `MAI-UI-8B`, and `MAI-UI-2B`. Third-party sites that describe Z-Image-Edit as "released" are describing a model card, not a checkpoint. **Z-Image is not Qwen-Image.** Both are Alibaba, both are Apache 2.0, both generate images — and they are separate model lines from separate teams. On arena.ai, [Qwen-Image](https://howaiworks.ai/models/qwen-image)'s `qwen-image-2512` outranks Z-Image-Turbo. ## Capabilities - **8-step generation.** Z-Image-Turbo "matches or exceeds leading competitors with only 8 NFEs (Number of Function Evaluations)." The quick-start sets `num_inference_steps=9`, which the model card notes "actually results in 8 DiT forwards," and `guidance_scale=0.0` — Turbo runs without classifier-free guidance entirely. - **Bilingual text rendering.** Alibaba claims Z-Image "excels at accurately rendering complex Chinese and English text" inside generated images, and the report calls bilingual text rendering one of two areas where results "rival top-tier commercial models." - **Photorealism.** The second of those two areas. The blog phrases it as "Photography-level Realism." - **Sub-second latency.** On an enterprise H800. The consumer-GPU path is slower but fits in under 16 GB of VRAM. - **Undistilled base for builders.** The base Z-Image "preserves the complete training signal," "supports full Classifier-Free Guidance," "responds with high fidelity to negative prompting," and is positioned as "a good base for LoRA training, structural conditioning (ControlNet) and semantic conditioning." - **Prompt Enhancer.** An optional reasoning pass that Alibaba says lets the model "transcend surface-level descriptions and tap into underlying world knowledge." - **Reward post-training.** Turbo is the only variant with an RL stage, via the DMDR method (Distribution Matching Distillation meets [Reinforcement Learning](https://howaiworks.ai/glossary/reinforcement-learning), arXiv 2511.13649). ## Technical Specifications - **Hugging Face repositories**: `Tongyi-MAI/Z-Image-Turbo`, `Tongyi-MAI/Z-Image` - **Architecture**: Scalable Single-Stream Diffusion [Transformer](https://howaiworks.ai/glossary/transformer) (S3-DiT). Text, visual semantic tokens, and image VAE tokens "are concatenated at the sequence level to serve as a unified input stream, maximizing parameter efficiency compared to dual-stream approaches." - **Parameters**: 6B (vendor figure, referring to the DiT). The base repo's transformer shards total 12.31 GB in bf16 — consistent with ≈6.15B parameters. - **Transformer config**: 30 layers, model dim 3840, 30 attention heads, 2 refiner layers, 16 input channels, QK-norm enabled - **Text encoder**: a separate `Qwen3ForCausalLM` (hidden size 2560, 36 layers), *not* counted in the 6B - **VAE / scheduler**: `AutoencoderKL` / `FlowMatchEulerDiscreteScheduler` - **Sampling — Turbo**: 8 steps, no CFG (`guidance_scale=0.0`), no negative prompting, not fine-tunable - **Sampling — base**: 28–50 steps, `guidance_scale` 3.0–5.0, negative prompting supported, fine-tunable - **Resolution**: the base model card recommends **512×512 to 2048×2048 total pixel area, any aspect ratio**. The Turbo quick-start generates at 1024×1024. Alibaba publishes no separate resolution range for Turbo, so this page does not state one. - **VRAM**: <16 GB for Turbo (vendor claim) - **Download size**: 20.55 GB (base, bf16 transformer); 32.90 GB (Turbo, fp32 transformer) - **Training cost**: 314K H800 GPU hours, "approx. $630K" - **License**: Apache 2.0, on both repositories and the GitHub LICENSE file Alibaba's own Model Zoo table grades the variants rather than ranking them, and the trade-off is explicit: Turbo scores **"Very High"** visual quality but **"Low"** diversity and **N/A** fine-tunability; the base model scores **"High"** quality with **"Medium"** diversity and **"Easy"** fine-tunability. ## Use Cases - **Local, offline generation on a consumer GPU** — the model's reason for existing. No API key, no per-image cost, no content pipeline leaving your machine. - **Signage, posters, and UI mockups with embedded text** — bilingual Chinese/English rendering is one of the two capabilities Alibaba claims parity with commercial models on. - **LoRA and ControlNet development** — use the *base* checkpoint, not Turbo. Alibaba explicitly marks Turbo as non-fine-tunable. - **High-throughput batch generation** — 8 forward passes per image with no CFG means roughly one-sixteenth the compute of a 50-step CFG sampler. - **Commercial products under a clean license** — Apache 2.0 with no revenue cap, unlike [Stable Diffusion 3.5](https://howaiworks.ai/models/stable-diffusion)'s $1M Community License threshold. - **Distillation and few-step research** — Decoupled-DMD (arXiv 2511.22677) and DMDR (arXiv 2511.13649) are published methods with open reference weights to study them against. ## Performance / Benchmarks Two independent arenas rate this model, and **both are far less flattering than Alibaba's marketing.** Note that **arena.ai** (the former LMArena) and **Artificial Analysis** are different organisations running different Elo scales; their numbers are not comparable to each other. ### arena.ai text-to-image leaderboard (fetched 2026-07-08; vote cutoff July 5, 2026) | Rank | Model | Org | Elo | License | |---|---|---|---|---| | 1 | gpt-image-2 (medium) | OpenAI | 1385.04 | Proprietary | | 2 | muse-image | Meta | 1279.57 | Proprietary | | 3 | reve-2.0 | Reve | 1270.92 | Proprietary | | 4 | gemini-3.1-flash-image-preview | Google | 1270.07 | Proprietary | | 5 | mai-image-2.5 | Microsoft AI | 1257.47 | Proprietary | | … | *(ranks 6–11: Google ×3, OpenAI, xAI, Ideogram)* | | | | | 12 | qwen-image-2.0-pro | Alibaba | 1192.55 | Proprietary | | 33 | qwen-image-2512 | Alibaba | 1126.86 | Apache 2.0 | | **48** | **z-image-turbo** | **Alibaba** | **1080.92** ±5.9 | **Apache 2.0** | Z-Image-Turbo ranks **#48 of 72 models**, on 19,719 votes. Two facts follow from the table above, and both cut against the hype: 1. **The entire top 11 is US labs** — OpenAI, Meta, Reve, Google, Microsoft AI, xAI, and Ideogram. The best Chinese entry is Alibaba's *proprietary* qwen-image-2.0-pro at #12. On the closed image frontier, Chinese labs are not leading; on open weights, they are. 2. **Z-Image-Turbo is not even Alibaba's best Apache-2.0 image model on this board.** `qwen-image-2512`, also Apache 2.0, sits fifteen places higher. ### Artificial Analysis text-to-image leaderboard (fetched 2026-07-08) | Rank | Model | Org | Elo | Open weights | |---|---|---|---|---| | 8 | Cosmos3-Super-Text2Image (agentic) | NVIDIA | 1,226 | Yes | | **69** | **Z-Image Turbo** | **Alibaba** | **1,104** | **Yes** | | 104 | Z-Image Base | Alibaba | 1,034 | Yes* | Artificial Analysis's own FAQ names **Cosmos3-Super-Text2Image (agentic)** as the current open-weights leader, at #8 overall. \* Artificial Analysis records no open-weights URL against its Z-Image Base row, which appears to be a gap in its data rather than a licensing fact — the checkpoint is public and Apache 2.0. ### The vendor's own claims, and what happened to them Alibaba's News log still carries, dated **2025-12-08**: *"Z-Image-Turbo ranked 8th overall on the Artificial Analysis Text-to-Image Leaderboard, making it the #1 open-source model!"* That was true when written. **It is no longer true.** Seven months later the same leaderboard places Z-Image Turbo at #69. Alibaba's other performance claim — *"state-of-the-art results among open-source models"* — is sourced to **Alibaba AI Arena**, an arena Alibaba itself operates. We attempted to verify it: `aiarena.alibaba-inc.com` renders entirely client-side and returns **zero model names** in its HTML. It is not first-party verifiable, and it is a vendor grading its own homework — which is the ordinary condition of a [state-of-the-art claim](https://howaiworks.ai/glossary/state-of-the-art-model) rather than an exception. We reproduce the claim and decline to endorse it. **What the numbers actually support:** Z-Image is a strong, fast, permissively licensed model that fits on a consumer GPU. It is the most-downloaded text-to-image model of Chinese origin on Hugging Face. It is not the highest-rated image model, nor the highest-rated open one. ## Limitations - **Popular ≠ best.** #48 of 72 on arena.ai and #69 on Artificial Analysis. Downloads measure accessibility, not quality, and Z-Image's download lead is largely a function of its size and license. - **Turbo cannot be fine-tuned, guided, or negatively prompted.** Alibaba's own table lists fine-tunability as **N/A**, CFG as ❌, negative prompting as ❌, and diversity as **"Low"** — distillation collapses seed-to-seed variety. Anyone building LoRAs must start from the base model, which is 44× less downloaded. - **Two of four advertised variants have never shipped.** Z-Image-Omni-Base and Z-Image-Edit have been "To be released" since November 2025, and the repository has been untouched since February 9, 2026. - **The 16 GB VRAM figure is not the download size.** The Turbo repository ships an fp32 transformer and totals **32.90 GB** on disk. - **"6B" excludes the text encoder.** The pipeline also loads a Qwen3 model with a 2560 hidden size and 36 layers. Real memory and disk footprints are larger than "6B" implies. - **All quality claims are vendor claims.** The report's "performance comparable to or surpassing that of leading competitors" and the "state-of-the-art among open-source models" line rest on Alibaba's own arena, which cannot be independently read. Both independent arenas disagree. - **The Hugging Face card tags the model `language: en`** despite the bilingual claim, and Alibaba publishes no text-rendering accuracy benchmark for either language. - **No first-party hosted API.** Every live inference provider on Hugging Face is a third party, flagged `isModelAuthor: false`. ## Pricing & Access ### Self-hosting **Free, under a real open-source licence.** Apache 2.0 on `Tongyi-MAI/Z-Image-Turbo`, on `Tongyi-MAI/Z-Image`, and in the GitHub LICENSE file. No revenue cap, no field-of-use restriction, no geographic exclusion. Commercial use, modification, and redistribution are permitted. Training data is not released, so this is open weights *plus* an open licence — not open data. ### Hosted access Hugging Face's inference-provider mapping lists three live hosts for Turbo. **All three are third parties**; Alibaba operates none of them, and each is marked `isModelAuthor: false`: | Provider | Provider model ID | |---|---| | fal.ai | `fal-ai/z-image/turbo` (base: `fal-ai/z-image/base`) | | Replicate | `prunaai/z-image-turbo` | | WaveSpeed | `wavespeed-ai/z-image/turbo` | Replicate's PrunaAI deployment publishes per-image pricing tiered by resolution: **$0.0025** up to 0.5 MP, **$0.005** up to 1 MP, **$0.01** up to 2 MP. This is a third-party price for a third-party deployment, not an Alibaba rate card. **Alibaba publishes no first-party hosted price for Z-Image that we were able to verify**, so this page does not state one. Free demos run on Hugging Face Spaces (`Tongyi-MAI/Z-Image-Turbo`, `Tongyi-MAI/Z-Image`) and on ModelScope. ## Ecosystem & Tools - **[diffusers](https://github.com/huggingface/diffusers)** — `ZImagePipeline` is merged into the main branch (PRs #12703 and #12715). The canonical inference path. - **ComfyUI** — native support; `ZImage` classes appear in `comfy/supported_models.py`. A third-party [ComfyUI-ZImageLatent](https://github.com/HellerCommaA/ComfyUI-ZImageLatent) node supplies the official resolutions. - **[stable-diffusion.cpp](https://github.com/leejet/stable-diffusion.cpp)** — a pure C++ engine that the project says runs Z-Image "on machines with as little as **4GB** of VRAM," across CUDA and Vulkan. - **[DiffSynth-Studio](https://github.com/modelscope/DiffSynth-Studio)** — LoRA training, full training, distillation training, and low-VRAM [inference](https://howaiworks.ai/glossary/inference). - **[vllm-omni](https://github.com/vllm-project/vllm-omni)** and **[SGLang-Diffusion](https://github.com/sgl-project/sglang)** — production serving paths. - **[Cache-DiT](https://github.com/vipshop/cache-dit)** — nearly 4× speedup on 4 GPUs "with negligible precision loss"; also covers Z-Image-ControlNet. - **[Candle](https://github.com/huggingface/candle)** — Hugging Face's Rust ML framework supports Z-Image. - **[MeanCache](https://github.com/UnicomAI/MeanCache)** — training-free acceleration, up to 3.7× (China Unicom). ## Community & Resources - [Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer](https://arxiv.org/abs/2511.22699) — the technical report (v1 2025-11-27, v5 2026-07-06); source of the 6B, 314K GPU-hour, $630K, and <16GB figures - [Tongyi-MAI/Z-Image on GitHub](https://github.com/Tongyi-MAI/Z-Image) — 11,697 stars, Apache 2.0, the News log with both release dates - [Tongyi-MAI/Z-Image-Turbo on Hugging Face](https://huggingface.co/Tongyi-MAI/Z-Image-Turbo) — Turbo weights, Model Zoo table - [Tongyi-MAI/Z-Image on Hugging Face](https://huggingface.co/Tongyi-MAI/Z-Image) — undistilled base weights - [Z-Image project blog](https://tongyi-mai.github.io/Z-Image-blog/) — the official landing page - [Decoupled DMD: CFG Augmentation as the Spear, Distribution Matching as the Shield](https://arxiv.org/abs/2511.22677) — the 8-step distillation method - [Distribution Matching Distillation Meets Reinforcement Learning](https://arxiv.org/abs/2511.13649) — DMDR, Turbo's RL post-training - [arena.ai text-to-image leaderboard](https://arena.ai/leaderboard/text-to-image) — where Z-Image-Turbo ranks #48 - [Artificial Analysis text-to-image leaderboard](https://artificialanalysis.ai/image/leaderboard/text-to-image) — where it ranks #69 - Compare with [Stable Diffusion 3.5](https://howaiworks.ai/models/stable-diffusion), [Qwen-Image 2.0](https://howaiworks.ai/models/qwen-image), [HunyuanImage 3.0](https://howaiworks.ai/models/hunyuan-image), [HiDream-O1-Image](https://howaiworks.ai/models/hidream), and [Seedream 5.0](https://howaiworks.ai/models/seedream) ## Frequently Asked Questions ### When was Z-Image released? The project's own News log dates Z-Image-Turbo to November 26, 2025 and the undistilled Z-Image base model to January 27, 2026. The Hugging Face repositories were created slightly earlier — 2025-11-25 and 2026-01-23 respectively — because the weights were uploaded before the public announcement. ### Is Z-Image open source? Yes, genuinely. Both `Tongyi-MAI/Z-Image-Turbo` and `Tongyi-MAI/Z-Image` carry the `license:apache-2.0` tag on Hugging Face, the GitHub repository ships a verbatim Apache License 2.0 file, and arena.ai independently records the license as Apache 2.0. There are no field-of-use or geographic restrictions. Training data is not released. ### How big is Z-Image? Alibaba describes it as a "6B" parameter family, and that figure refers to the diffusion transformer alone. The transformer weight shards total 12.31 GB in bf16 on the base repo — consistent with roughly 6.15B parameters. The pipeline additionally loads a separate Qwen3 text encoder, so a full base download is 20.55 GB and the full Turbo download is 32.90 GB. ### Can Z-Image run on a consumer GPU? Yes — that is the entire point of the model. The technical report claims "compatibility with consumer-grade hardware (<16GB VRAM)" for Z-Image-Turbo. The project also links stable-diffusion.cpp, which reports running Z-Image on machines with as little as 4 GB of VRAM. ### Is Z-Image the best open image model? No, and this is the most important correction on this page. Z-Image is the most-downloaded text-to-image model of Chinese origin, not the highest-rated. On arena.ai's text-to-image leaderboard it sits at #48 of 72 with an Elo of 1080.92 — below Alibaba's own Apache-2.0 `qwen-image-2512` at #33. On Artificial Analysis it is #69 with an Elo of 1,104, while NVIDIA's Cosmos3-Super-Text2Image leads open-weights models at #8. ### Wasn't Z-Image ranked #1 open-source on Artificial Analysis? It was, in December 2025. The project's News log still advertises "ranked 8th overall on the Artificial Analysis Text-to-Image Leaderboard, making it the #1 open-source model!" dated 2025-12-08. That claim is now stale on both counts: as of July 8, 2026 Artificial Analysis places Z-Image Turbo at #69 overall, and the open-weights leader is NVIDIA's Cosmos3-Super-Text2Image (agentic) at #8. ### How popular is Z-Image on Hugging Face? Very. The Hugging Face API reports 933,580 downloads for `Tongyi-MAI/Z-Image-Turbo` in the trailing 30 days and 6,660,069 all-time, with 4,944 likes. That makes it the third most-downloaded model carrying the `text-to-image` tag, behind only Stable Diffusion v1.5 (1,718,458) and SDXL base 1.0 (1,439,261). ### Are Z-Image-Edit and Z-Image-Omni-Base available? No. Both are described at length in the model card, the blog and the technical report, but the Model Zoo table has marked them "To be released" since launch. The Tongyi-MAI Hugging Face organisation contains only four repositories, and neither variant is among them. Do not plan around them. ### What is the difference between Z-Image and Z-Image-Turbo? Turbo is the distilled, RL-post-trained variant: 8 sampling steps, no classifier-free guidance, no negative prompting, and — per Alibaba's own table — "Low" output diversity and no fine-tunability. The base Z-Image is the undistilled transformer: 28–50 steps, CFG 3.0–5.0, working negative prompts, and it is the checkpoint intended for LoRA training and ControlNet work. ### How much does Z-Image cost? Self-hosting is free under Apache 2.0. Alibaba publishes no first-party hosted API price that we could verify. Hugging Face's inference-provider mapping lists three live third-party hosts — fal.ai, Replicate and WaveSpeed — none of them operated by the model author. Replicate's `prunaai/z-image-turbo` publishes $0.0025 per image up to 0.5 MP, $0.005 up to 1 MP, and $0.01 up to 2 MP. ## Related ### Related models - [Stable Diffusion 3.5](https://howaiworks.ai/models/stable-diffusion) - [Qwen-Image 3.0](https://howaiworks.ai/models/qwen-image) - [HunyuanImage 3.0](https://howaiworks.ai/models/hunyuan-image) - [HiDream-O1-Image](https://howaiworks.ai/models/hidream) - [Seedream 5.0](https://howaiworks.ai/models/seedream) - [Qwen3.8-Max](https://howaiworks.ai/models/qwen) --- Source: https://howaiworks.ai/models/z-image — HowAIWorks.ai