Photos & Videos Ai Hunks - Diffusion generated stunning hotties

Thank you! I wish I could share/engage more on here but this thread is just mostly people reposting stuff.

Refiner, no. Only the base model works well with a refiner in my experience. I use hires fix at a low denoising strength to achieve a better result with any XL model I've played with including mine.

Yes, you could definitely train a checkpoint with them, but it would require some hefty GPU specs and a long training run to get a great model. How many do you have? You could also try training just a LoRA or Lycoris with them, this straight version was trained on 20,000 images and accomplishes a lot of poses (this LoRA is actually the base for most of the NSFW XL models, and even mine has this in the DNA somewhere which is one of the reasons for watermarks). I am not sure how he captioned images, but that would be the most daunting task involved.

I wouldn't be surprised if I had 20K images from different studios (SC, CF, ChaosMen, Men At Play, Titan Media, Falcon Studios, etc.). I enjoy collecting the stylized images the studios release. I honestly haven't found much use for them aside from the LoRAs I've created. I'd like to put the mass amount of data to good use.

I thought due to the volume a checkpoint would make sense, but I'd love if I could achieve the same with just a LoRA since that is all I've worked with at this point.
 
  • Like
Reactions: Trendyauto