Single-Stream Diffusion Transformer Architecture (6B)
Unlike traditional dual-stream models, Z-Image Turbo unifies text prompt tokens and visual image tokens into a single transformer stream.
- 6 billion parameters (6B) with dense cross-modal semantic attention
- Distilled rendering in just 8 function evaluations (8 NFEs) instead of 40-50 steps
- Natural rendering of skin texture, fabrics, glass reflections, and studio lighting