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.

fal/flux-1-schnell

API

FLUX.1 [schnell] is Black Forest Labs' 12B Apache-2.0 model, distilled with latent adversarial diffusion distillation to generate in 1-4 steps. Released August 2024 as the fast, permissively licensed member of the FLUX.1 family; benchmarked here on fal.

Maker
Black Forest Labs
Family
FLUX.1
Model Size
12B
high
Cost
$0.003 / megapixel
medium
Run Target
fal/fal-ai/flux/schnell
Effective Request
image_size: square_hd · num_images: 1 · num_inference_steps: 4 · seed: 7
38.3
Overall
37%
Capability
39.5
Est. Preference
71
Pass
121
Fail
4.1s
Avg Latency
3.0s
Min Latency
8.0s
Max Latency
Text Rendering33%Spatial Reasoning44%Human realism19%Truthfulness33%Professional Studio59%Graphical design33%Preference40%Latency59%

All 192 generations

Text Rendering33%

Typography Style67%

Writing accuracy25%

Spatial Reasoning44%

Attributes Binding56%

Compositionality67%

Counting22%

Negation33%

Relative Position42%

Scale & Proportions44%

Human realism19%

Faces & Expressions17%

Full Body8%

Hands42%

Multi-Subject0%

Truthfulness33%

Photorealism0%

Physics & Reflections42%

World Knowledge33%

Professional Studio59%

Camera & Lighting67%

Color Precision58%

Photorealism33%

Graphical design33%

Data Visualisation0%

Layout & Design22%

Style Diversity50%