ImageBench

ImageBench V1 —

192 evaluations across 6 categories

Benchmark V1 verdicts are produced by VLM judges and can contain mistakes. Treat PASS/FAIL labels as machine-assisted assessments, and inspect the images yourself. Learn more about the methodology.

Generation Details

Source-backed model context, size, cost, and request settings for this ImageBench V1 run.

local/z-image-turbo-6b

Local

Z-Image Turbo is a distilled 6B text-to-image model from Tongyi-MAI, built for efficient few-step generation with strong photorealism, bilingual English/Chinese text rendering, and practical local deployment.

Maker
Tongyi-MAI / Alibaba
Family
Z-Image
Model Size
6B
high
Cost
local run; no API price
not_applicable
Run Target
z-image-local/z-image-turbo
Effective Request
size: 1024x1024
Sources
Z-Image-Turbo model cardTongyi-MAI Z-Image repository
49.2
Overall
48%
Capability
49.9
Est. Preference
93
Pass
99
Fail
18.1s
Avg Latency
12.0s
Min Latency
26.8s
Max Latency
Text Rendering60%Spatial Reasoning49%Human realism31%Truthfulness33%Professional Studio82%Graphical design50%Preference50%Latency15%

All 192 generations

Text Rendering60%

Typography Style100%

Writing accuracy50%

Spatial Reasoning49%

Attributes Binding78%

Compositionality56%

Counting33%

Negation22%

Relative Position67%

Scale & Proportions33%

Human realism31%

Faces & Expressions67%

Full Body8%

Hands8%

Multi-Subject50%

Truthfulness33%

Photorealism0%

Physics & Reflections58%

World Knowledge17%

Professional Studio81%

Camera & Lighting83%

Color Precision75%

Photorealism100%

Graphical design50%

Data Visualisation0%

Layout & Design33%

Style Diversity75%