<h1 id="what's-new:">What's new:</h1>I've added more training data focusing on back views and side views. Due to some adjustments in the training parameters, the generalization is a bit limited this time. <h3 id="for-better-results-try-the-following-prompts-:">For better results, try the following prompts :</h3>For front view:<ul><li>foreshortening,view from below,footstomp</li><li>splitt0es, foot focus, feet</li></ul>For side view:<ul><li>low angle from side,worm's-eye view,foreshortening,footstomp</li><li>s1t,leftvi3w, crossed legs,feet,5toes,sitting（1：1）</li><li>s1t,leftvi3w, crossed legs,feet, barefoot,sitting（4：3）</li><li>s1t,leftvi3w, crossed legs,feet,5toes,sitting（3：4）</li></ul>For back view:<ul><li>lying down,lying on bed</li><li>1iedown,backvi3w, back view, soles（1：1）</li><li>1iedown, backvi3w, back view, 5toes（4：3）</li><li>1iedown,lying down, backvi3w, back view, soles（3：4）</li></ul>You might find that the model responds differently depending on the canvas size. I highly recommend experimenting with different prompt combinations—especially when switching between various aspect ratios—to find the best results for your specific composition.

Files with SHA256: 18ce5ea9498ce2d89b32aad681d427a8b9f8e39819e3c3380fb941238ff1adc3