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/kromoss-dmd

Local

Kromoss is lodestones' Apache-2.0 full fine-tune of MOSS-Image-Base at 1024 px (704k images), a drop-in replacement for MOSS-Image-SFT with the same Qwen3.5-9B text encoder and FLUX.2 small-decoder VAE. This entry adds the 8-step DMD turbo LoRA.

Maker: lodestones
Family: Kromoss (MOSS-Image)
Model Size: 5B
medium
Cost: local run; no API price
not_applicable
Run Target: gx10/kromoss-dmd
Effective Request: response_format: b64_json · size: 1024x1024 · num_inference_steps: 8 · guidance_scale: 1 · seed: 7
Sources: lodestones/Kromoss model card MOSS-Image project page

51.8

Overall

63%

Capability

41.0

Est. Preference

120

Pass

72

Fail

18.5s

Avg Latency

18.1s

Min Latency

19.1s

Max Latency

All 192 generations

Text Rendering67%

Typography Style100%

Writing accuracy58%

Spatial Reasoning60%

Attributes Binding78%

Compositionality78%

Counting33%

Negation44%

Relative Position67%

Scale & Proportions56%

Human realism45%

Faces & Expressions75%

Full Body0%

Hands83%

Multi-Subject0%

Truthfulness44%

Photorealism33%

Physics & Reflections50%

World Knowledge42%

Professional Studio93%

Camera & Lighting92%

Color Precision100%

Photorealism67%

Graphical design83%

Data Visualisation33%

Layout & Design100%

Style Diversity83%