Only use girls for training____________A composite concept, but a little overtrained, but even so, it still cannot retain the details and effects well.Originally, there were three levels: small gaping anus, gaping anus, huge gaping anus, but the level distinction was not good enough, and secondly, the huge gaping anus effect could not be effectively presented, because the original composite and processed pictures were far exaggerated countless times to the point where the buttocks almost disappeared, leaving only an anus.Now it seems that only "gaping anus" has an effect, and the overtraining factor is recommended to be 0.8~0.9.________I am preparing for the concept training of lora on lumina2, but the immaturity of the related tools is a big problem, and I am looking forward to the release of Nvidia's cheap 24GB VRAM graphics card. It would be great if it could be cheaper... so that I can use 1536 or even 2048 resolution training.____Some of my more powerful concepts are more or less affected by problems and are not released, such as a girl sitting on a boy in the crowd on the tram, anal x-ray effects are displayed on the abdomen, and there is a huge penis.There are also some effects such as internal human organ perspective, or all the way through, and even the ability to allow guro, including the operability of organs such as stomach, esophagus, large intestine, small intestine, and rectum. However, due to the complexity of the concept and the fact that TE has been trained to dizziness and cannot be correct, appropriate simplification can be relatively stable output but the details are messy...The girl is inserted into the fountain and sprays water and shows x-ray effects inside her body, the monster's huge tongue and body penetrate the girl's body, and come out of her mouth or ass, etc. There are countless strange effects, but even with hundreds of samples, it is still not stable.TE limits the concept's play and structural stability, but without training the model, it is impossible to understand...

Files with SHA256: f94a87962acd278c031a4b040c2eccd431ec9a8681bdc1ece4eb5be74578171a