Krea 2: Updated the dataset to upscaled and enhanced 2K 2048px to see if it would make a noticeable difference than 1024px / 1536px, and it did. The image gallery contains one-shot examples using a base Krea 2 workflow (No Refusal LoRAs not necessary). At first I didn't notice too much of a different until I set the image HxW values with one value being 2048px or so, then I was able see the difference.

I made a minor change to the previous captioning that should make spicy generations a bit better . I combed through the captioning and corrected some arousal state mismatches that were there before. So specifying erect or flaccid in prompt should yield more successful generations.

Flux 2 9B v1: I expanded the roster a bit to include an Arab / Persian American featured character, who I'm naming Omar Awad. The body variant I was going for was lithe build with a dominant pectoral / over-developed pecs physique (you know the type that always skip leg days!) -- as always, he comes fully meat-equipped. I was on the fence with cut or uncut but went with circumcised since I'm pretending he's American.

I asked Omar if he minded that I created him without a dong turtleneck given he was modeled after Arabic and/or Persian characteristics and he was like "NaN"

Anyways, the images in the gallery are based off standard BFL Flux 2 Klein 9B model (I use quantiezed GGUF Q5_K_M) and for these examples stuck with Euler a 4 step count and that's it. No artificial additives like GMOs or SNOFs LoRas added .. just as the good AI Lord(s) made 'em.

Enjoy.