V2 : it tried again from scratch , for those who download v1 or / and will v2 i wouldnt mind earing about you and your tried , if it does change thing some or not , it wasnt same image use, it was well scan by blip / deepboroo for taging , it seem this time to take into account the weights if you put any after the trigger word , and seem to render term like anime uselessmight try again later on , and if any got some info or so , i did include the dataset use to trainthanks for all V1 Hi, here embedding / textual inversion i tried to train in stable diffusion, not reallty sure if i didnt put not enough image , or put bad setting, but i thought it give something a bit like nearly text generation, not sure if of help to any one

strange result

training starting from scratch 2500 steps and about 70 epoch

Files with SHA256: e71939c768c5d8627f59cc2ba5113e868d2c19210513a9be57a2893f60d77d01