Z-Image-Base

High Performance: The model family (including the "Turbo" version) has ranked as a top open-source model on the Artificial Analysis Leaderboard, rivaling closed-source giants.

Efficiency: It is built on a 6-billion parameter architecture that can run on consumer-grade hardware (requiring as little as 16GB of VRAM).

Bilingual Text Rendering: It is particularly skilled at rendering accurate text within images in both English and Chinese.

Versatility: The "Base" model is intended as a foundation for fine-tuning, while other variants like Z-Image-Turbo (for sub-second generation) and Z-Image-Edit (for instruction-based image editing) cater to specific use cases.

Architecture: It utilizes a "Scalable Single-Stream DiT" (S3-DiT) architecture, which treats text and image data as a unified stream to maximize efficiency.