*I have updated my training method and am now (finally) got the CAME optimiser working on my AMD system, so expect a lot more posts going forward (and smaller file size, GO TEAM RED!!). From the Wiki...Nothing from the Wiki anymore, I'm just going to be focusing on my Loras. You know how you got here... * I train my characters on lots of tags rather than using the common "absorption" method. This is because, sometimes, i wish to be very specific, what is ripped, gloves or no gloves etc. If you are just here to generate some quick NSFW images of one of your preferred waifus, you can simply put the character name and 999 times out of 1000 get a good generation. Maybe uniform, maybe costume, who knows, but if you also have pussy, anus, breasts in your prompt, pay it little never-mind... Prompt: toga himiko Then (if you want): blonde hair, messy hair, double bun, serafuku, cardigan (whatever you like, examples attached, can be adjusted accordingly, for example "open cardigan") All my character loras should work well together. I have updated my training method so prompts should be a little simpler. Try to keep tags as minimal as possible to avoid bleeding (or just use in-painting, whatever). My example images (attached) use minimal character features in prompting. Trust in the Lora.

Files with SHA256: 528c5a7025fe40d4c14bbcff5653c4b8599a8d4a31357e7c3eae62acfbaef795