{
 "title": "GPT Image 2 VIP native 4K test, LaoZhang API",
 "generated": "2026-11-09",
 "timezone": "UTC+08:00",
 "endpoint": "POST https://api.laozhang.ai/v1/images/generations",
 "token_group": "default",
 "run": "One request from one machine; same prompt as the 2026-11-08 test in ../gpt-image-4k-2026-11/.",
 "files": [
  {
   "image": "gpt-image-2-vip-3840x2160-high.png",
   "sha256": "67e51da017b16f365abc900e2586026e5b23fe11133944d5a9c8951945915d26",
   "bytes": 11442376,
   "pixels": [
    3840,
    2160
   ],
   "request": "gpt-image-2-vip-3840x2160-high.request.json",
   "request_sha256": "c4eac27dd60a8db59e39380256f0e67c45b63959743671ed77f774805467bacd",
   "response": "gpt-image-2-vip-3840x2160-high.response.json",
   "response_sha256": "3bb2ee6afab7aef3da021ceeebe882edc8b11fe9437e464397f5237d41751ec2",
   "model": "gpt-image-2-vip",
   "size": "3840x2160",
   "quality": "high",
   "http_status": 200,
   "attempts": 1,
   "seconds": 89.3,
   "request_sent": "2026-11-09T09:59:31+08:00",
   "response_received": "2026-11-09T10:01:01+08:00",
   "charged_usd": 0.03,
   "charge_logged_at": "2026-11-09T10:00:59+08:00",
   "response_usage_output_tokens": 13342,
   "psnr_round_trip_db": {
    "whole_image": 36.4,
    "most_detailed_quarter": 33.8,
    "script": "check_4k.py"
   },
   "controls_psnr_round_trip_db": {
    "enlarged_from_1920x1080": {
     "whole_image": 48.2,
     "most_detailed_quarter": 45.5
    },
    "enlarged_from_1536x864": {
     "whole_image": 49.5,
     "most_detailed_quarter": 47.9
    }
   },
   "c2pa": {
    "claim_generator": [
     "Adobe Firefly"
    ],
    "signer": "Adobe Inc.",
    "c2patool_failures": [
     "signingCredential.untrusted"
    ]
   }
  }
 ],
 "notes": [
  "Image: the bytes returned by the API, base64-decoded and saved unchanged. Verify with: shasum -a 256 <file>.",
  "Response file: the base64 image is replaced by its byte count and sha256. The usage block is kept as returned; it is not a billing record, because gpt-image-2-vip is billed at a flat $0.03 per call. Its 13,342 output tokens match the count OpenAI publishes for a 3840x2160 high GPT Image 2 image.",
  "charged_usd is the per-call charge in the account's usage log (15,000 quota units = $0.03).",
  "PSNR: python3 check_4k.py <file> (Pillow LANCZOS, half size and back). It reports the whole image and the most detailed quarter, because flat areas raise the whole-image score: this file scores 36.4 dB whole but 33.8 dB on its most detailed quarter. Controls made by shrinking this file to 2K or 1.5K and enlarging it back score 45.5-47.9 dB on the detailed quarter. The 2026-11-08 manifest used ImageMagick for its figures, which come out about 1 dB lower than this script on the same files.",
  "C2PA: the PNG carries a Content Credentials manifest whose claim generator is Adobe Firefly, signed by Adobe Inc.; the 2026-11-08 GPT Image 2.5 and 2.6 -vip files are signed by OpenAI OpCo, LLC with claim generator OpenAI Media Service API. c2patool reports signingCredential.untrusted because its default trust list does not include the certificate."
 ]
}