<h2 id="purpose">Purpose</h2>Limited to one specific pose: <ul><li>Female standing, facing forward, with knees and toes turned inwards, AKA a knock-knees and pigeon-toe.</li><li>This isn't a good lora. I'll make a v1.0 some day. But it does this one pose, so I'm sharing it.</li></ul>Gallery images: <ul><li>left side = lora, right side = no lora</li><li>Workflows are embedded for the Krea gallery</li></ul><h2 id="krea2-version">Krea2 version</h2>Usage: <ul><li>Use strength is 0.6 to 0.8. There's no trigger. </li><li>Even though it wasn't captioned, adding "her knees are pressed together and feet touching" to your prompt will enhance the pose. </li><li>Don't use "shy pose" in your prompt unless you want the hands near the face, and face turned away. That effect isn't caused by this lora, but from the Krea2 base.</li></ul>Limitations, sadly: <ul><li>No males</li><li>No Side-profile</li><li>No Sitting or lying down poses</li><li>Causes slight de-aging</li></ul>Training method: <ul><li>I used LLM captions and 20 images: 10 closeups of legs+feet, and 10 full body with all faces obscured.</li><li>I used 1,200 steps, starting with super high learning rate and sharply descending, then trail and error block tuning. </li><li>Normal learning rate with more steps would have been better!</li></ul><h2 id="zit-version">ZiT version</h2>Usage: <ul><li>Start your prompt with the full phrase, "A woman is standing in a shy pose with her toes pointing together." </li><li>Use strength 1.3.</li></ul>Limitations, sadly:<ul><li>Only one person</li><li>No males</li><li>No side profile</li><li>No Sitting or lying down poses</li><li>Impacts faces and other aspects of the image</li></ul>Training method: <ul><li>Trained for $2 on a service that had no settings</li><li>I used 20 images: legs and feet only (I learned later this is a bad technique)</li></ul>

Shy pose with pigeon toe

Krea2 version, also a bad lora. Sorry! I'll make a good one some day.

Files with SHA256: 6ae24e5529de3c88ee9e0c93546af0efee0fe8f0e4fea5794821e42124f0aca2