{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreib3axv7spls77f22rxfxhwj4kjjlmg7qsqos5wk57ecsu5ktjh3ke",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3meoev4kjkyx2"
  },
  "path": "/t/why-does-my-vision-fine-tuning-job-keep-failing/1371510#post_7",
  "publishedAt": "2026-02-12T15:09:22.000Z",
  "site": "https://community.openai.com",
  "tags": [
    "documented"
  ],
  "textContent": "Hey Folks,\n\nApologies for the inconvenience and I agree that the behaviour is very vague to understand the issue. Vision fine-tuning is currently documented for gpt-4o-2024-08-06 only. If you’re trying to run “vision FT” on another base model, you can hit validation failures during image processing. The vision FT pipeline is strict about image inputs. Make sure every image is:\n\nJPEG/PNG/WEBP\n\nRGB/RGBA\n\n<= 10MB\n\nand if using URLs, they must be publicly accessible (no auth, no expiring signed URL issues, no blocked access). Base64 format: If embedding images, the docs expect data URLs (not raw base64 blobs). I tested a few and it seems to work without any issues. Please let me know, if that helps. Thank you!",
  "title": "Why does my vision fine-tuning job keep failing?"
}