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/iris-3b

Local

Iris-3B by Sperid Labs is a 3B-parameter Apache-2.0 diffusion transformer that generates directly in pixel space: no VAE and no latent, with 16x16 RGB patches, 2D RoPE and a frozen Qwen3-VL-4B text encoder read at eleven layers. Trained progressively up to about one megapixel. This entry is the base model sampled with the author's defaults (100 DPM-Solver++ steps, CFG 3.0).

Maker: Sperid Labs
Family: Iris-3B
Model Size: 3B
high
Cost: local run; no API price
not_applicable
Run Target: gx10/iris-3b
Effective Request: response_format: b64_json · size: 1024x1024 · num_inference_steps: 100 · guidance_scale: 3 · seed: 7
Sources: speridlabs/iris-3b model card Iris-3B: Going Beyond the Latent with Pixel-Space Diffusion (arXiv 2610.09450)speridlabs/iris-3b (GitHub)