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/hidream-o1-image-8b

API

HiDream O1 Image is an 8B pixel-native text-to-image model built on a Pixel-level Unified Transformer with no external VAE or separate text encoder, open-sourced under the MIT licence in May 2026. This is the undistilled checkpoint.

Maker: HiDream.ai
Family: HiDream O1
Model Size: 8B
high
Cost: $0.0105 / image at 1024x1024
high
Run Target: fal/fal-ai/hidream-o1-image
Effective Request: image_size: [object Object] · num_inference_steps: 50 · guidance_scale: 5 · enable_safety_checker: false · output_format: png
Sources: HiDream O1 Image (8B) on Hugging Face HiDream O1 Image (8B) on fal HiDream-O1-Image paper (Pixel-level Unified Transformer)

57.7

Overall

53%

Capability

62.8

Est. Preference

101

Pass

91

Fail

19.4s

Avg Latency

6.1s

Min Latency

158.1s

Max Latency

All 192 generations

Text Rendering73%

Typography Style100%

Writing accuracy67%

Spatial Reasoning56%

Attributes Binding67%

Compositionality67%

Counting44%

Negation56%

Relative Position67%

Scale & Proportions33%

Human realism33%

Faces & Expressions50%

Full Body0%

Hands58%

Multi-Subject17%

Truthfulness37%

Photorealism0%

Physics & Reflections50%

World Knowledge33%

Professional Studio67%

Camera & Lighting75%

Color Precision75%

Photorealism0%

Graphical design67%

Data Visualisation0%

Layout & Design56%

Style Diversity92%