Attention ☞ Recommended parameters:
Sampler: Euler
Scheduler: simple
Note:
Steps: 9+
Online - Steps: 10+
CFG ：1
Z-Image-Turbo-Aesthetic Realism-Asian Realism-FP8 is a model fine tuned based on Z-Image-Turbo-FP8, which is trained on Xiaohongshu and fused with Asian realism
🚀 Z-Image Turbo - a refined version of Z-Image, with only 8 NFEs (functional evaluation times) to match or surpass leading competitors. It is available on enterprise grade H800 GPU ⚡ Sub second level inference delay ⚡ ️, and comfortably installed in 16GB video memory consumer devices. It excels in realistic image generation, bilingual text rendering (in English and Chinese), and strict teaching adherence.
Realistic image quality: Z-Image-Turbo provides powerful realistic image generation while maintaining excellent aesthetic quality.
Accurate bilingual text rendering: Z-Image-Turbo excels at accurately rendering complex Chinese English texts.
Hint Enhancement and Reasoning: Hint Enhancers endow models with reasoning abilities that go beyond surface descriptions and touch upon potential world knowledge.
Creative Image Editing: Z-Image Edit demonstrates a profound understanding of bilingual editing guidance, enabling imaginative and flexible image transformations.
Model Architecture
We adopt a scalable single stream DiT (S3 DiT) architecture. In this setting, text, visual semantic symbols, and image VAE symbols are concatenated at the sequence level as a unified input stream, maximizing parameter efficiency compared to the dual stream method.
Performance:
According to Elo based human preference assessment (AI), Z-Image-Turbo demonstrates high competitiveness compared to other leading models and has achieved state-of-the-art results in open-source models.
Description
Attention ☞ Recommended parameters:
Sampler: Euler
Scheduler: simple
Note:
Steps: 9+
Online - Steps: 10+
CFG ：1
Z-Image-Turbo-Aesthetic Realism-Asian Realism-FP8 is a model fine tuned based on Z-Image-Turbo-FP8, which is trained on Xiaohongshu and fused with Asian realism
🚀 Z-Image Turbo - a refined version of Z-Image, with only 8 NFEs (functional evaluation times) to match or surpass leading competitors. It is available on enterprise grade H800 GPU ⚡ Sub second level inference delay ⚡ ️, and comfortably installed in 16GB video memory consumer devices. It excels in realistic image generation, bilingual text rendering (in English and Chinese), and strict teaching adherence.
Realistic image quality: Z-Image-Turbo provides powerful realistic image generation while maintaining excellent aesthetic quality.
Accurate bilingual text rendering: Z-Image-Turbo excels at accurately rendering complex Chinese English texts.
Hint Enhancement and Reasoning: Hint Enhancers endow models with reasoning abilities that go beyond surface descriptions and touch upon potential world knowledge.
Creative Image Editing: Z-Image Edit demonstrates a profound understanding of bilingual editing guidance, enabling imaginative and flexible image transformations.
Model Architecture
We adopt a scalable single stream DiT (S3 DiT) architecture. In this setting, text, visual semantic symbols, and image VAE symbols are concatenated at the sequence level as a unified input stream, maximizing parameter efficiency compared to the dual stream method.
Performance:
According to Elo based human preference assessment (AI), Z-Image-Turbo demonstrates high competitiveness compared to other leading models and has achieved state-of-the-art results in open-source models.