Z-Image-Turbo

Z-Image-Turbo is the model that Alibaba's Tongyi lab released first from its Z-Image family — before the foundation model it was distilled from. The weights went up on Hugging Face on 25 November 2025 and the code repository followed a day later, both under Apache 2.0, which is unusual for a Chinese image generator of this standing: most ship under bespoke licences with usage clauses attached. The technical idea is speed without a quality collapse. Where a diffusion model normally needs dozens of denoising passes, Turbo is trained to finish in eight — which Alibaba reports as sub-second generation on an H800 accelerator, and which brings a 1024×1024 image within reach of a consumer graphics card with 16 GB of memory. Two of the lab's own papers describe how: Decoupled-DMD, which separates the two mechanisms that make distribution-matching distillation work, and DMDR, which folds reinforcement learning into that distillation. The architecture behind both is what Alibaba calls a single-stream diffusion transformer — text, visual semantics and image tokens concatenated into one sequence of 30 layers rather than split across two towers. What Turbo is good at and what it gives up are both on the record. It renders Chinese and English text inside an image accurately, holds photorealism well, and follows instructions closely; in exchange it takes no negative prompts, ignores guidance scale, produces markedly less variation between seeds than the undistilled base, and Alibaba lists it as not suited to fine-tuning. For that, the foundation model Z-Image arrived two months later. The performance claim — best among open-weights models on an Elo leaderboard of human preferences — comes from Alibaba's own AI Arena, so it is the vendor's measurement rather than an independent one. Traffic, at least, is not in doubt: 681,390 downloads from the official repository in the thirty days to 17 September 2026, plus 7.76 million more from a community repackaging of the same weights for ComfyUI.

▸Videos

No videos yet.

▸Reviews