Trained on 38 photos on @senchan_25 Twitter account. (If any legal issues are involved, please delete this model immediately.)The model has a certain bias and prefers to generate side-face photos (this may be because the training data is mainly based on such photos).This is just a model for my experiment, and there will be no updates in the future.<h3>How I train Lora</h3>1. Collect the girl's photos: by hand or crawl from SNS.<ul><li>Weibo: I use <a target="_blank" rel="ugc" href="https://github.com/dataabc/weibo-crawler">weibo-crawler.</a></li><li>Instagram: <ul><li>chrome extension: <a target="_blank" rel="ugc" href="https://chrome.google.com/webstore/detail/download-albums-for-insta/inkncgklbglecgdlcpfpajejocdpbpbd">https://chrome.google.com/webstore/detail/download-albums-for-insta/inkncgklbglecgdlcpfpajejocdpbpbd</a> (suggested by balkanboy)</li><li>command line: <a target="_blank" rel="ugc" href="https://instaloader.github.io/">https://instaloader.github.io/</a> (suggested by kiwidaci52)</li></ul></li><li>I don't know how to crawl Twitter. If anyone knows the method and is willing to share it, I would be very appreciate.</li></ul>2. Use <a target="_blank" rel="ugc" href="https://colab.research.google.com/github/Linaqruf/kohya-trainer/blob/main/kohya-LoRA-finetuner.ipynb#scrollTo=gPgBR3KM6E-Z">Linaqruf's Colab notebook</a> to train Lora. (Many thanks to his adaptation work and kohya-ss for his training script.)May be <a target="_blank" rel="ugc" href="https://rentry.org/59xed3">this document</a> can be helpful. I learned a lot from it.Hope more and more Textual Inversion and Lora for Instagram/Twitter/Weibo Girl.

Files with SHA256: a25c71c7a50d3910ab5d2a0d6d793ecee20e9d8a3e155ab3ce5f379ef599e1e8