NOTE: /user/freckledvixon and /user/Wiggin has an excellent lora with these similar shoes, but Im not a fan of the ankle straps/anklets in both the Loras and whenever I try to prompt out the straps/anklets on the ankles I get some ugly artifacts that linger around the ankles from the training data. Also, both these LORAs seem to not have included 'ass/rear shots" of a female in these heels, so the heels always comes out warped and mutated, something ive included in my dataset.NOTE: If you're getting mainly pure feet pic crops in your generations, try removing the "female feet" portion of the trigger word, that seems to trigger close up shots of feet.NOTE: Toes will almost always come out badly, thats just a Stablediffusion 1.5 problem that will never be fixed in 1.5. Not even Loras dedicated to perfect toes seem to fix the problem.These shoes go by many names. Platform heel, Mules, Slip-ons, Slides, etc.I have a fetish for this particular type of heel and Ive been waiting a long time for someone to make a lora for it, but nobody did so here I am. Ive included the training data, mostly open sourced from public domains across the web like google image, Amazon store, etc. I would personally benefit vastly if anyone decided to pick it up and improve upon it. If anyone has a problem or claim to any pics in the training data, please contact me and I will remove it and retrain the LORA.Seems like a strength of 0.7 to 0.8 works the best. Anything over 0.9 seems to burn/fry the picture. Like everything in StableDiffusion 1.5, the further out you cropped a person in an image the more details you lose. I like full body shots always so Im having a terrible time trying to get it to work for me, mainly the toes, but thats just a 1.5 issue. 1.5 has always had extreme difficulty getting toes and fingers right. I cant use SDXL or I would make a LORA for it, SDXL melts my pc(potato PC).

Plat mules

alpha/rushed version 0.1. Will improve in the future. 1. Caption data is not perfect. In the next version I intend to caption rear facing feet/heels and the colors(clear, black, red, pink, etc), but thats going to take quite a long time since there are 800 pics in the training data. 2. Next version I will cull low quality images in training data, and cull heels that do not match with overall data. 3. Since this is an early alpha I have put a lot of restrictions on it, no onsite generation, need to be logged in to download, etc etc etc. I will remove most restrictions when Ive felt like Ive perfect it, or when im satisfied with the lora being the final version.

Files with SHA256: 36d01003c2954e89bf6c433ffb1c64715dc622148963a15f5cb1dfb27e7e0952