girl green shirt black hair walking beach night sky (stars:0.7) hair

I've been wondering about the impact of prompt syntax on model generation. Does a comma have a tangible effect? Does repeating the same word have an effect, and if so is it different from weighting the words?

I ended up making this prompt, and then passed it through a number of different models. For each model, I ran a set of 6 to 8, chose 1, and then upscaled it.

I had a hard time deciding which of the Wonton images to choose, so I upscaled the lot and included the batch image.

I noticed that in general, using 4x-UltraSharp tended to make the image feel more dramatic at higher denoising levels, while ERSGAN_4x provided softer textures, which were sometimes preferable, depending on the scene.