Z-Image-Turbo
6B single-stream diffusion transformer — sub-second photorealistic generation with bilingual text rendering.
About
Z-Image-Turbo is the distilled, speed-optimized member of Alibaba Tongyi-MAI's Z-Image (Zaoxiang) family, released in November 2025 under Apache 2.0. It is a 6-billion-parameter single-stream diffusion transformer that matches or exceeds leading competitors with only 8 function evaluations (NFEs), delivering sub-second inference latency on H800 GPUs while fitting in 16GB of consumer VRAM. Its practical strengths are photorealistic image generation, bilingual English/Chinese text rendering, and robust instruction adherence. It ships in the diffusers ecosystem with an online demo on Hugging Face Spaces and ModelScope.
The family includes three siblings for other jobs: Z-Image, the base foundation model tuned for aesthetics, diversity, and controllability; Z-Image-Omni-Base, a generation-and-editing foundation checkpoint aimed at community fine-tuning; and Z-Image-Edit, fine-tuned for precise natural-language image editing. The technical report (arXiv 2511.22699) documents the single-stream DiT design. Performance claims are the provider's; the architecture, license, checkpoint, and demo are independently confirmable.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: Verified via Hugging Face model card (Tongyi-MAI/Z-Image-Turbo): 6B single-stream DiT, distilled to 8 NFEs, sub-second H800 latency, 16GB VRAM, bilingual EN/ZH text rendering, diffusers ecosystem.; Weights confirmed downloadable on HF (ungated); repo created 2025-11-25; 5.5k likes; technical report arXiv 2511.22699; sibling variants (base, omni-base, edit) documented in card.; Speed/quality-vs-competitors claims are provider's; architecture, license, checkpoint, and live demo independently confirmable.. License: Apache-2.0 (model card frontmatter).