Z-Image

byAlibaba / Qwen team

Z-Image is an efficient, open-foundation AI image-generation model built on a compact 6 billion-parameter Scalable Single-Stream Diffusion Transformer (S3-DiT) architecture. It’s designed to balance high-quality, photorealistic output with fast inference and low compute requirements, making advanced image generation accessible even on consumer-grade hardware with ~16 GB VRAM. Z-Image delivers strong realism, accurate bilingual text rendering (e.g., English and Chinese), and robust prompt adherence, while variants like Z-Image-Turbo focus on ultra-fast generation with sub-second latency.