{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreidfippb6qtdyjpoi72zgnp7pnyylnul2lhugdvvcgrnszkrekipcq",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mjeqw7emec62"
  },
  "path": "/t/total-ai-beginner-with-a-25-year-photography-archive-is-this-useful-for-training/175157#post_6",
  "publishedAt": "2026-04-13T09:36:53.000Z",
  "site": "https://discuss.huggingface.co",
  "textContent": "rcopeck:\n\n> I’ve heard people say AI images aren’t quite right because they are training on other AI images—do these older files act as a 'clean’ baseline or beginning?\n\nAI that makes images is getting better and better each month. You should try a newer model now and generate some images. People now have the right number of fingers, the right number of arms, and clipping (one object intersecting another which wouldn’t happen in the real world) is seen less and less with newer models.\n\nThere are some amusing AI images from 2025. In one example, a game controller was plugged into a bag of chips. And the girl was only wearing one sock for some reason. There are whole videos on Youtube about AI image blunders",
  "title": "Total AI beginner with a 25-year photography archive—is this useful for training?"
}