Sorry, too much whiskey, I guess.. have changed that now. Thx :-)

Hi, thank you so much for your feedback. I always appreciate it when someone takes the time to let me know how my Loras are working for them, even if, as in your case, they didn’t work so well. A 1 in 20 success rate is, of course, not acceptable. Could you also let me know how you’re generating the images? Are you using Forge or ComfyUI with my workflow that’s integrated into each image?

@rollpollution193 @rollpollution193 Oh, thank you very much. I'm glad to hear that. I hope I can explain this in my terrible English: Like everyone else here, I have no idea what Flux expects from me as input for a LoRA or how exactly I should formulate it, because it's all so new. At this early stage, it's still a lot of trial and error. For my LoRAs, I've taken a slightly different approach and haven't relied solely on BLIP, GIT, WD14, or any of the other image captioning algorithms to generate the text for each image, but instead, I've chosen my own method. I took a WD14 text that's close to the concept (with a general threshold of 0.05) and simply added a longer text in front, which, in my opinion, best reflects what's happening in the image. Sometimes this works better in my LoRAs, sometimes not so much. It probably depends on the subject matter, because Flux, in my view, was trained perfectly for any possible output, but in no way for the human body or human anatomy. Sometimes you have to explain things to Flux in very long sentences, like when it comes to topics like semen, in detail, so it knows what I'm even asking for. Sometimes that doesn't work at all either. Like I said, trial and error. I hope this helps. Why my LoRAs work better than others, I really can't say. It makes me happy, but only God (or the guys over at Black Forest Labs) know why.