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.

replicate/google/imagen-4-ultra

API

Imagen 4 Ultra is the highest-quality tier of Google DeepMind's Imagen 4 family, targeting maximum detail and the strongest prompt alignment for brand-critical or production work. It supports up to 2K resolution with the family's typography improvements.

Maker
Google DeepMind
Family
Imagen 4
Model Size
not disclosed
unknown
Cost
$0.06 / image
high
Run Target
replicate/google/imagen-4-ultra
Effective Request
aspect_ratio: 1:1
62.0
Overall
62%
Capability
61.9
Est. Preference
119
Pass
73
Fail
14.6s
Avg Latency
9.0s
Min Latency
50.0s
Max Latency
Text Rendering73%Spatial Reasoning70%Human realism33%Truthfulness37%Professional Studio89%Graphical design83%Preference62%Latency21%

All 192 generations

Text Rendering73%

Typography Style100%

Writing accuracy67%

Spatial Reasoning70%

Attributes Binding78%

Compositionality100%

Counting56%

Negation67%

Relative Position75%

Scale & Proportions44%

Human realism33%

Faces & Expressions33%

Full Body17%

Hands58%

Multi-Subject17%

Truthfulness37%

Photorealism0%

Physics & Reflections33%

World Knowledge50%

Professional Studio89%

Camera & Lighting92%

Color Precision83%

Photorealism100%

Graphical design83%

Data Visualisation0%

Layout & Design89%

Style Diversity100%