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/nucleus-image-17b-a2b

Local

Nucleus-Image is a 17B-parameter sparse mixture-of-experts diffusion transformer from NucleusAI, activating roughly 2B parameters per forward pass to target strong quality with efficient inference.

Maker
NucleusAI
Family
Nucleus-Image
Model Size
17B total, ~2B active
high
Cost
local run; no API price
not_applicable
Run Target
nucleus-local/nucleus-image
Effective Request
size: 1024x1024 · num_inference_steps: 30
Sources
Nucleus-Image model cardNucleus-Image paper
36.6
Overall
42%
Capability
31.5
Est. Preference
80
Pass
112
Fail
39.1s
Avg Latency
36.5s
Min Latency
39.8s
Max Latency
Text Rendering53%Spatial Reasoning47%Human realism21%Truthfulness26%Professional Studio67%Graphical design46%Preference32%Latency0%

All 192 generations

Text Rendering53%

Typography Style100%

Writing accuracy42%

Spatial Reasoning47%

Attributes Binding67%

Compositionality78%

Counting11%

Negation33%

Relative Position58%

Scale & Proportions33%

Human realism21%

Faces & Expressions42%

Full Body8%

Hands8%

Multi-Subject33%

Truthfulness26%

Photorealism33%

Physics & Reflections42%

World Knowledge8%

Professional Studio67%

Camera & Lighting67%

Color Precision75%

Photorealism33%

Graphical design46%

Data Visualisation0%

Layout & Design22%

Style Diversity75%