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/tinydit-256

Local

TinyDiT-256 is a 210M-parameter text-to-image DiT trained from scratch on a single GPU in 3.5 days (400k steps, 4.2M images at 256px). It uses rectified flow with a frozen FLUX.2 autoencoder and flan-t5-base text encoder, and renders only small trained shapes such as 256x256. Weights are CC BY-NC-4.0 (non-commercial).

Maker: Ivan Mikhnenkov (independent)
Family: TinyDiT
Model Size: 210M
high
Cost: local run; no API price
not_applicable
Run Target: gx10/tinydit-256
Effective Request: response_format: b64_json · size: 256x256 · num_inference_steps: 20 · guidance_scale: 4 · seed: 7
Sources: ivanmikhnenkov/tinydit-256 model card ivanmikhnenkov/tinydit (GitHub)