<a rel="ugc" href="https://www.instagram.com/sophieraiin/">Sophie Rain</a><a target="_blank" rel="ugc" href="https://www.instagram.com/sophieraiin/"> </a>is an instagram and onlyfans model. This is a 120 step TI from a 15 image dataset using <a target="_blank" rel="ugc" href="https://civarchive.com/user/JernauGurgeh/models">JernauGurgeh's</a> method of training embeddings. This was trained primarily using close-up or chest-up images, with 5 of them showing most of her body.<h2 id="heading-27">How to create a good prompt using this TI</h2>I recommend Jernau's checkpoint and starting with this phrase specifically"photo of s0phierain-120, a woman"Extra descriptors include nose ring, necklaces, and open mouth.If you have feedback about any limitations or suggestions to improve this, let me know. Personally, I find dreambooth fine-tuned models merged with your favorite checkpoint to be more effective than LoRAs or embeddings, but I figured I would share this since I finally found success with training a TI.

Files with SHA256: 51cf9e193377e5d8f31ed231b0c956237b9d2760024627c00e24113e63d0ad22