Nice plot!

Usually you want multiple rolls, cause visual performance heavily varies with the seed. You roll at least 5 pics for each model to get the idea what it tends to draw.

Also, different models have different "languages", despite striving for naturalness. Which model/arch suits you the best cannot be answered with a uniform prompt, but that's probably out of scope of any straightforward benchmark.

"Tired of scrolling through endless model rankings without truly SEEING the results, I built https://ImageBattle.ai to visually compare AI-generated images from multiple models. Because seeing is believing."