{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreibx62nopwbjnktxe7sdvybvktceqaujamjyu3az6lhstqdr6sdia4",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mhcxg2hhgf22"
  },
  "path": "/t/about-traning-lora-for-z-image-turbo/173911?page=2#post_24",
  "publishedAt": "2026-03-18T06:48:39.000Z",
  "site": "https://discuss.huggingface.co",
  "tags": [
    "@John6666"
  ],
  "textContent": "I have now successfully trained my LoRa character. To those reading this discussion, I repeat: there are no other settings that affect the training process as much as the dataset, caption, and step. There’s also the learning rate, but according to my tests, this shouldn’t be more than 0.0001 for ZIT anyway.\n\n@John6666, I’m now going to start generating the NSFW dataset. But it seems the values ​​we used for the LoRa character will be insufficient for this side. I’d like to hear your opinion. According to my research, since genital areas contain a lot of detail, rank 16 and around 40 images will be insufficient. What are your thoughts on this?",
  "title": "About traning LoRa for Z Image Turbo"
}