I'm tired of bad experiences from low-quality FLUX Loras so I'm on a journey to make my own robust &amp; awesome ones.It is honestly early-access worthy, but I'm giving this one out to the community for free. I might reconsider that for further Loras as it was a lot of work to carefully prepare the dataset, editing manually most masks then training it carefully for as many epochs as I could before risking overfitting. README : Features of this Lora (aka why it's different from some poor loras I've tried before) : - It cannot alter other Loras faces AT ALL because the dataset was prepared using loss masks on ALL the faces, including males ones - It was trained on 80 porn pictures of white girls, mainly without the male face in the original pictures. Only small to medium boobs. No big boobs, no saggy boobs, no chubby/fat girls, mostly nice &amp; slim body type. Disclaimer : If you don't like that, just don't download it and make your own. I made this primarly for me.- Dataset contains mainly front close-ups but seems to be able to do (with proper prompting) full body/side doggy/plowpose/etc. YMMV. - It is so strong I've found out it can even work with weights as low as 0.1 - 0.2 - It was captioned using long, natural language captions similar to what flux loves - All watermarks where previously removed carefully - Apparently it can do anime as well although it wasn't trained on it at all Important Guidelines - The trigger word is "DFCpose" - You MUST use natural language prompt as if you were describing fully the doggy pose facing camera yourself, like you had no Lora. The efficiency of the lora depends on it, as it was captioned this way- A lot of captions contained "woman/couple engaging in sexual act facing camera" so I use everytime - You SHOULD play with the weights of this lora and your character face Lora to find the optimal balance. This Lora can work at anything from 0.1 weight to 1. - I recommend you push your character's face lora to the maximum it can handle (1 - 1.5) to be sure you don't end up with a "mild face" (see below the Drawbacks section). For example this lora 0.3 and your face lora 1.3. Play with combinations for each lora.- If generating anime, emphasize clearly on it with "Anime drawing of", "anime character girl", etc. Consider using an anime checkpoint. You might end up with the guy not being anime, emphasize on him as well. Drawbacks - No dicks. Due to the pose, there just wasn't any in the dataset. - It doesn't, and cannot alter the face BUT using a 2nd lora can "dilute the face" in the output, especially if your face lora is weak, resulting in a "mild face" which looks like your character but not as good as if your face lora was alone. Not much we can do except play with the weights. - Your character face can appear occasionally on the guy behind's face as they were not part of the training as well - You will encounter occasionnaly some anatomy problems or weird poses, success rate is not 100% and relies heavily on your own prompting of the pose Examples prompts (that I've tested and work - also design your own) :<ul><li>Full-body view of nude couple engaging in sexual act facing camera, woman is standing bent forwards, leaning forwards facing viewer, with a naked man exactly behind her with his legs across her ass and his hands grabbing her</li><li>View of nude couple facing camera, engaging in sexual act, woman is standing on her knees looking at viewer, with the man behind grabbing her ass</li><li>Full-body view of woman, engaging in sexual act, standing on her knees looking at viewer, having sex with a man standing behind her</li><li>Full-body view of woman, facing camera, engaging in sexual act, crawling on her hands and knees, looking at viewer, having sex with a man standing behind her and grabbing her hips</li><li>Side Full-body view of naked woman engaging in sexual act, facing camera, crawling on her hands and feet, a naked man standing behind her with his legs around her ass and grabbing her </li></ul>Thanks for reading. Happy fluxing ! Please post your creations :-)

Files with SHA256: 5fab74761511a9c5d2f878c8adb7bb25e4ac71e499364cede524a45812e62f4a