I originally made this for dipping (feet sliding in and out of shoes) but it works pretty well for adding general foot and toe movements to Wan generations.Obviously works best with shoes on. Trained with flip flops but works for other styles as well.Trained on base Wan 2.2 A14B I2V, but all preview videos were made with <a target="_blank" rel="ugc" href="https://civarchive.com/models/2053259/wan-22-enhanced-nsfw-or-camera-prompt-adherence-lightning-edition-i2v-and-t2v-fp8-gguf?modelVersionId=2540892">taek's Enhanced NSFW model</a>. PLEASE note this lora just does foot movements. all the NSFW preview videos here are thanks to the NSFW checkpoint!High noise strength 0.3~1.0 (0.5 is a good starting point)Low noise strength 1.0Higher strength on the high noise will reduce/slow overall motion with the lightning lorasPhrases I use most often (try other ones too. get creative.)<pre><code>d1pp1ng. she is wearing [shoes].</code></pre><pre><code>she slides her feet repeatedly back and forth within her shoes.</code></pre><pre><code>her toes flex and curl as they fill the sandal straps.</code></pre>I also usually add...<pre><code>her feet are moving continuously throughout the entire scene.</code></pre>...toward the end of the prompt, because otherwise she sometimes stops moving her feet halfway through.The "physics" is generally pretty good but sometimes feet clip through straps. Higher weight will improve the physics but also slow the speed. so you need to experiment with weight.Technobullshit:<ul><li>Started with a single dipping video, 4 clips. I trained a lora on that, then used it to generate 500 clips. ~100 of those turned out decent.</li><li>~90% of the clips were feet only / knees-down, ~10% were full body.</li><li>I made several copies of the clips, adjusted to various speeds, trimmed, mirrored everything, and wound up with 496 clips for training this lora.</li><li>Trained both high and low on the same set, both at 512x512 res (nominal), high noise to 1788 steps, low noise to 3576. rank 16 alpha 16. musubi-tuner on 5090.</li></ul>

Files with SHA256: 108076ab378e1c6b2605c792f366f8a8c81935ee39b0480c46515d55fffcaacd