How to detect Midjourney images
Midjourney images have a distinctive look and leave specific metadata traces. Knowing what to look for — and why pixel-level detectors sometimes miss them — makes identification much more reliable.
Visual characteristics of Midjourney images
Midjourney produces highly stylized, painterly results. Common tells include:
- Hyper-realistic skin with an airbrushed, pore-free quality unlike any camera.
- Ornate, symmetrical compositions — Midjourney tends to centre subjects and add elaborate backgrounds.
- Unusual bokeh: soft, dreamy out-of-focus areas that don't match the apparent lens.
- Warped or absent text: signs, labels and badges are the most reliable visual tell in Midjourney v5 and earlier; v6 improved text but still produces subtle errors.
- Hands and fingers: Midjourney has improved significantly since v5, but merged or extra fingers still appear under unusual angles.
None of these alone is conclusive. Recent versions (v6 and Niji 6) are substantially harder to spot by eye.
Metadata traces
Midjourney images served directly from their CDN carry minimal EXIF, but the XMP block often contains the prompt text or a Midjourney job ID. Upload the file to AiPicDetect and open the metadata panel — if you see an xmp:Description or Iptc4xmpCore field referencing "Midjourney", that is near-conclusive evidence.
Midjourney does not (as of 2026) sign C2PA content credentials on standard image exports, so the absence of a C2PA block tells you nothing about origin.
Using AiPicDetect on Midjourney images
The underlying model was trained on a broad mix of AI generators. Midjourney tends to score High-confidence AI on direct exports. However:
- Screenshots and re-uploads reduce the score — JPEG compression artifacts add "real-photo" signal.
- Heavily edited Midjourney images (composited, run through photo editors) may score Uncertain.
- A Medium or Uncertain score on a visually suspicious image should be treated as a hint, not clearance.
For the most reliable result, upload the original file rather than a screenshot. If you only have a screenshot, run all four checks from the main guide.
Quick reference
| Signal | Weight for Midjourney |
|---|---|
| XMP prompt or job-ID field present | Strong |
| High-confidence AiPicDetect score on original file | Strong |
| Painterly skin, centred composition, unusual bokeh | Supporting |
| Warped text or off-count fingers | Supporting (weaker on v6) |
| Screenshot, no original available | Inconclusive without metadata |
Try the detector on your image →
Cập nhật lần cuối 2026-09-13 · mã nguồn trên GitHub