Generator guide

How to spot Midjourney images

Midjourney is one of the most used image generators and one of the fastest moving. What was a reliable tell in 2024 — polished skin, painterly light, six fingers — has mostly disappeared in current versions. Recognition shifts from style to measurement.

Why style no longer suffices

Older Midjourney versions had a recognisable look: dramatic lighting, smooth textures, a faint oil-paint feel. Current versions render grain, blur and imperfect compositions that resemble ordinary photography. Anyone watching for the old look misses the new output.

What still works

• Text and logos: even new versions break on coherent letters at zoom.
• Repeated structures: tiles, wood grain, background crowds — patterns repeating just a bit too regularly.
• Physical consistency: hands holding objects without contact, reflections that do not add up, jewellery changing between photos of the same 'person'.
• Over-smooth distribution: a series where every image is equally successful, without the failures a real shoot always contains.

What our detector measures

On 150 Midjourney images from August 2026 the current lane drew a strong warning on 81.3% (95% interval 74.4–86.7%). On older Midjourney generations in the same setup it was 67.1%. The newest output is not invisible, but it is the lane's weakest spot — and we publish that figure separately instead of averaging it away.

How to use this

If you suspect Midjourney origin: check the image with the image detector, then zoom into text and hands, and run a reverse image search. A strong score plus a physical inconsistency is a reason to ask follow-up questions; neither is proof on its own. All figures, samples and limits are on the accuracy page.