Nice plot!<p>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.<p>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 <a href="https://ImageBattle.ai" rel="nofollow">https://ImageBattle.ai</a> to visually compare AI-generated images from multiple models. Because seeing is believing."