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/mgflow-flux2-klein-4b

Local

MGFlow is an open-weight one-step text-to-image model post-trained from FLUX.2 [klein] 4B. It generates a 512x512 image in a single network evaluation with no classifier-free guidance, trained by matching real and generated features in frozen representation spaces with Gaussian-mixture statistics (MGFlow-KL). This entry is the joint image-text checkpoint, submitted by the authors for evaluation. Apache-2.0 checkpoints, MIT code.

Maker: Shi Haoyang et al. (MGFlow authors)
Family: MGFlow
Model Size: 4B
high
Cost: local run; no API price
not_applicable
Run Target: gx10/mgflow-flux2-klein-4b
Effective Request: response_format: b64_json · size: 512x512 · num_inference_steps: 1 · guidance_scale: 0 · seed: 7
Sources: shy0423/MGFlow (checkpoints)Unifying Distributional Training for One-Step Visual Generation (arXiv 2609.35763)shihaoyang0423/MGFlow (GitHub)