I took some time to properly train a Lora for close-up soles poses that focuses less on pointed toes (which can be done with the Foot Focus Helper). Because the dataset is smaller with many repeats, it's less versatile but it provides better generation aesthetics for anime feet.I recommend weights 0.5 - 0.8 (My favorite is 0.6) for all generation types (T2I and I2I), any higher and you'll get very distorted anatomy. Trial and error are your best friend here. Lower weights prevent overpowering other LORAs used with it. I use a CFG scale of 7, For inpainting, you'll want to make sure 'masked only' is set and specifying "(single foot)" in prompt when doing individual soles can help prevent distortion. Some generations may give 4 toes on a foot -- this can be resolved with 0.45-0.5 denoise strength inpainting.IMPORTANT NOTE: Specifying leg orientation (spread legs, legs up, legs together, sitting, etc.) helps greatly with avoiding anatomy error. If you don't specify, you'll likely end up with limbs in strange directions since the poses it was trained on vary.Recommended Keywords:Trigger Word: SolesMasterExample prompt header: ((spread legs, legs up, SolesMaster, feet, toes, soles, barefoot, foreshortening))<ul><li>Specify 'legs up', 'feet up' etc. as desired</li><li>Specify 'close-up' when inpainting. Inpaint feet individually with 'masked only' set</li><li>This Lora has various nail polish colors in the dataset, but its not explicitly trained to generate nail polish, keep this in mind if you want to include painted nails in your generation as it will take a few tries.</li></ul>As always thanks for supporting, Buzz tips are appreciated so I can train other anatomy LORAs for my own works, (This one used quite a bit), since I over-trained the first variant w/ too many repeats and had to change settings. Also, check out my <a target="_blank" rel="ugc" href="https://www.patreon.com/MizarAiArt">Patreon </a>if you feel like supporting :)

Files with SHA256: 8b956b7c6b2e46263eda5c4b96c4bb0edc97d4fa028e80763246836e32f5c621