Trying out various A.I. models by generating 3 images with the exact same prompt on all of them. To keep sizes down for the test images, I used a 512 x 512 pixel image size, which is the default for many models.
The prompt is “Cute petite redhead with freckles in a sexy pink bikini. Standing on a beach, watching a sunset in the distance. Blue sky with very few clouds. Seagulls in the distance. Some driftwood washed ashore.” On top of that, the selected template added the following to the prompt: “photography. f/2.8 macro photo, bokeh, photorealism“.
FLUX.2 Klein 4B is a relatively small model, 2.6 GB. Here are three examples:
Dreamshaper 8 is a somewhat larger model (5.5 GB) but the results compared to FLUX are disappointing, even if it is more than 10 times faster. As always, larger doesn’t always mean better.
Qwen Image 2512 (Q8_0) is a rather large model, almost 22 GB, here’s one image. I’m not going to post more right now, because this single one took 10 hours to make because I still didn’t manage to get my GPU working and this is all running on CPU only.
I tried the same prompt on an old Stabile Diffusion v1.5 engine. Results were… unexpected…
Leave a Reply