Thank you! Is it the homofidelis sd 1.5 v4.0? Also what prompts + negative prompts did you use for this image?
Yes, that's it, except I'm using the v5 version currently. Here are my parameters:
Realistic, candid photo of a handsome, lean, fit, athlete, dark brown hair, pubic hair, pulling down briefs and holding his bulge, locker room, dim lighting, <lora:epi_noiseoffset2:1>, <lora:add_detail:1>
Negative prompt: bodybuilder, muscular, huge, CyberRealistic_Negative-neg
Steps: 36, Sampler: DPM++ 3M SDE, Schedule type: Karras, CFG scale: 5, Seed: 2030414283, Size: 512x768, Model hash: 8391342199, Model: homofidelis_v50
I find that it sometimes takes a fair amount of experimentation, to tweak the various parameters until you hit that "magic combo" for the results you're after. I should also note that I often use ControlNet with Canny or Depth to guide the generation with an input image (usually one that I've found somewhere – sometimes in a thread here), and experiment with those parameters until I get something I like.