{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreiex32c2b7upqc6ii4pkytcl4ty3et7wtspkkksh625ejx6eyskhtu",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mki25fhmsld2"
  },
  "path": "/t/collection-of-gpt-image-generator-2-0-issues-bugs-and-work-around-tips-check-first-post/1379535?page=5#post_100",
  "publishedAt": "2026-04-27T11:32:17.000Z",
  "site": "https://community.openai.com",
  "tags": [
    "(click for more details)"
  ],
  "textContent": "The generator has multiple security layers. Both the text that is entered and a check of the finished image are used. So there can be several stages that filter the image. (At least simple trigger words are no longer used like in earlier generators.) But text meaning.\n\nYou often have to experiment a bit to describe the image in a way that doesn’t trigger directly. I would try to use terms related to fashion, clothing presentation, etc. And avoid: nudity, more then 1 person, interaction, etc. And also avoid brand names.\n\nTo find the trigger, use each phrase separately, or use a half-half search.\n\n* * *\n\nBefore two trigger words were snow-white and black panther. So the security is a bit smarter now.\n\nHere is a example from the past. “snow-white” and “black panther” wehre trigger words. So we used “pure white” and “black puma”. (But first we must spend time to find the triggers…) It is a bit the same on your case. Rephrase prompt parts witch could trigger a block, or a image result witch will be blocked.\n\nPrompt (click for more details)",
  "title": "Collection of GPT-image-generator 2.0 issues, bugs, and work-around tips (check first post)"
}