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/krea-2-turbo-no-filter

Local

Krea 2 Turbo (no filter) is the distilled Krea 2 Turbo checkpoint run with a diff-patch LoRA that disables the model's built-in content filter, so it completes prompts the stock model would refuse — benchmarked here for comparison against stock Krea 2 Turbo.

Maker
Krea AI
Family
Krea 2 Open-Source
Model Size
12B
high
Cost
local run; no API price
not_applicable
Run Target
gx10/krea2-turbo-lora
Effective Request
response_format: b64_json · size: 1024x1024 · num_inference_steps: 8 · guidance_scale: 0 · seed: 7
53.8
Overall
54%
Capability
54.0
Est. Preference
103
Pass
89
Fail
71.5s
Avg Latency
70.7s
Min Latency
72.6s
Max Latency
Text Rendering60%Spatial Reasoning61%Human realism21%Truthfulness41%Professional Studio82%Graphical design71%Preference54%Latency0%

All 192 generations

Text Rendering60%

Typography Style100%

Writing accuracy50%

Spatial Reasoning61%

Attributes Binding56%

Compositionality89%

Counting78%

Negation22%

Relative Position67%

Scale & Proportions56%

Human realism21%

Faces & Expressions67%

Full Body0%

Hands0%

Multi-Subject17%

Truthfulness41%

Photorealism33%

Physics & Reflections42%

World Knowledge42%

Professional Studio81%

Camera & Lighting75%

Color Precision83%

Photorealism100%

Graphical design71%

Data Visualisation0%

Layout & Design67%

Style Diversity92%