Z-Image-Turbo

Z-Image-Turbo is a lightweight text-to-image model specially designed for efficient generation, achieving performance comparable to top-tier competitors with just 8 sampling steps (8 NFEs).

Ultra-fast and low-power: The H800 delivers sub-second inference, easily compatible with consumer-grade GPUs such as the RTX 3060.

Top-notch image quality: With an FID score of 7.2 (better than SDXL), it supports photorealistic rendering and high-quality Chinese and English text rendering.

Proficient in Chinese: Built-in Qwen3-VL ensures an accuracy rate of up to 92% for Chinese instructions, perfectly capturing traditional Chinese artistic moods.

───────────────────────────────────────────────────────────────────

Core Capabilities

⚡ Ultra-lightweight

Inference speed improved by 60%. On an RTX 4090, generating a 512px image takes only 2.3 seconds; the H800 achieves sub-second response times.

🇨🇳 Proficient in Chinese semantics

Optimized based on Qwen3-VL, with accuracy surpassing Flux.2. It precisely understands complex Chinese concepts such as “the misty rain of Jiangnan.”

📷 Comparable to photographic quality

Delivering high precision with an FID score of 7.2, it features rich details and perfectly supports rendering both Chinese and English characters within the image.

🛠️ Strong ecosystem compatibility

Natively compatible with ComfyUI, supports Flash Attention acceleration, and complies with domestic data compliance requirements.

───────────────────────────────────────────────────────────────────

Case Studies