I was unsatisfied with the available loras centered around this concept (honestly there isn't any that solely focus on the futanari theme), even with detailed prompting the results are very hit or miss, so I made my own lora from a personally curated data set.The data set heavily focus on a side-view futanari with a large dark penis, so that will be the usual results. I have not tried my lora on realistic models so I don't know if it is good for that. This first verison is kind of unstable, it will sometimes give weird results, I suppose its the limits of SDXL/Illustrious model or my lora needs further fine tuning.If you want the same results as me use the wai nsfw illustrious merge from: <a target="_blank" rel="ugc" href="https://www.shakker.ai/modelinfo/0f204323a06f40e18f8ffc5b1813df5a/WAI-NSFW-illustrious-SDXL-NSFW-illustrious-SDXL-NSFW-illustrious-SDXL-NSFW-illustrious-SDXL?from=personal_page">LINK</a> I also like <a target="_blank" rel="ugc" href="https://civarchive.com/models/1717562/reedxxxillustrioussdxl">Reed_XXX_Illustrioust_SDXL</a> and it gives me good results.Recommended Lora strength 0.8~1, CFG 5~7, 896x1152, Euler or Euler A at 20~25 stepsFor best results start with these prompts: perfect lighting, highly detailed, (2D:1.1), (anime style:1.1),1girl, femdom, malesub, futanari, futa with male, futa on male, irrumatio, (faceless_male), best quality, masterpiece, absurdres, futa_face_fuck Update Version 2:After some testing version 2 seems to be much more stable, I have included a more diverse data set and did many tweaks during the training process. Following the same recommendations as V1 should yield good results. Always include the trigger word "Futa Face Fuck". In Version 2 if you are a degenerate and do not want your futanaris to have a dark penis make sure to include "dark penis" in the negatives.Optionally If you want to be fancy you can use Highres. fix at 0.25~0.3 denoise, upscale at 1.5~2, 20 hires steps and use the R-ESRGAN 4x+ Anime6B upscaler.

Futa Face Fuck

Updated with a more diverse data set and more fine tuning. Version 2.0 should gives more consistent results now.

Files with SHA256: 573624b583e7c311b797d616e033127f5f27d8baa003c68af978eccc7efe9ef9