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/sefi-image-5b-rl

Local

SeFi Image 5B RL is a locally fine-tuned text-to-image model produced by the SeFi image-generation fine-tuning pipeline and run on an NVIDIA DGX Spark. It is the ~5B reinforcement-learning-tuned variant of the SeFi Image family. It is not a publicly released hosted product; no external model card or citation is disclosed.

Maker
SeFi pipeline
Family
SeFi Image
Model Size
~5B
estimated
Cost
local run; no API price
not_applicable
Run Target
gx10/sefi-image-5b-rl
Effective Request
Effective request fields unknown
47.7
Overall
56%
Capability
39.1
Est. Preference
108
Pass
84
Fail
130.7s
Avg Latency
124.0s
Min Latency
284.8s
Max Latency
Text Rendering93%Spatial Reasoning63%Human realism26%Truthfulness22%Professional Studio93%Graphical design67%Preference39%Latency0%

All 192 generations

Text Rendering93%

Typography Style100%

Writing accuracy92%

Spatial Reasoning63%

Attributes Binding89%

Compositionality89%

Counting56%

Negation44%

Relative Position58%

Scale & Proportions44%

Human realism26%

Faces & Expressions50%

Full Body8%

Hands8%

Multi-Subject50%

Truthfulness22%

Photorealism33%

Physics & Reflections25%

World Knowledge17%

Professional Studio93%

Camera & Lighting92%

Color Precision92%

Photorealism100%

Graphical design67%

Data Visualisation0%

Layout & Design56%

Style Diversity92%