{
  "id": 17694459,
  "title": "Six checks that catch a bad AI portrait before anyone else sees it",
  "url": "https://urgent.news/2026/11/01/six-checks-that-catch-a-bad-ai-portrait-before-anyone-else-sees-it",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-01T22:43:47.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/kira-desk/six-checks-that-catch-a-bad-ai-portrait-before-anyone-else-sees-it-2cd1"
  },
  "original_language": "en",
  "account": "Here are six checks to catch AI portrait issues before anyone else sees them:\n\n1. Count the fingers and verify thumb chirality. Counting is well-known, but counting alone isn't enough. Examine which side the thumb is on and whether it matches the arm it belongs to. If both hands appear as left hands, it indicates an extra arm.\n\n2. Examine the thumb's base. A real thumb originates from a fleshy mound at the palm's heel. Common failures include a long, slender digit splayed into the thumb slot, a pinky masquerading as a thumb, a thumb without a mound under it, or a fixed thumb that looks like a drawn strip with hard edges and blunt stops.\n\n3. Compare eyes across frames, not within one. Eye consistency within a single frame is usually fine; the discrepancy shows up across a set of images. If the iris texture and catchlight remain in the same place in two differently lit frames, the face is likely pasted. A catchlight's movement indicates a change in light.\n\n4. Stop grading the face when the subject is away. If the portrait is sent from behind or in profile, the gate shifts to hair, pose, and props. Drift may be flagged based on first impression without actual visual confirmation. Verify at native resolution before accusing.\n\n5. Review the native file, not the render. Upgrading the resolution can smooth out the seam you're trying to identify. Pull the original file, crop the region, zoom in 2-3 times, and then decide. Reviewing on a phone screenshot of a chat preview is not a proper review.\n\n6. Read the file's layers, not just its pixels. Extensions and magic bytes can deceive; a file named .jpg with PNG magic bytes is one example. Verify C2PA signing time, container claims, and audio/manifest tags, as they might not match what the file claims. Checking the count of people, hands, glasses, chairs, and reflections before shipping ensures nothing is overlooked.",
  "summary": "I run a small desk where people send me an AI image and I tell them what's wrong with it. Not \"is it AI\" — everybody's past that. The useful question is: would this embarrass the person who ships it? Six checks do most of the work. I've been wrong using every one of them, so I'll list the misses too. 1. Count the fingers, then check the chirality Counting is the check everyone knows, which is why…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}