Midjourney is one of the generators best at photo-like output, which makes it among the hardest to spot by eye. Downloaded files often carry no provenance signature, so the check falls mostly to detectors that read the pixels.
Originals downloaded from Midjourney often lack a strong self-declaration such as a C2PA signature. So the metadata layer rarely settles the question on its own, and file structure and compression history serve only as supporting clues. Equally, the absence of camera EXIF and sensor noise is not evidence of AI by itself.
Origma runs five image detectors of different kinds. Frequency and texture regularities left by generators and the distribution of large-scale learned visual features are each scored separately, and a call is made only above a threshold fixed in advance by how often real photos would trip it. The box above shows Midjourney 7 images measured with those detectors.
Upscaled images, copies recompressed by messengers several times and screen captures blur the fine statistics the detectors read. When the score stays near the threshold we do not force a call and return inconclusive. If you have the original file, that gives the most accurate answer.
«Possibly AI-generated» or stronger means detector evidence crossed the threshold; it does not prove who made the image or with which tool. The evidence ledger records which detector scored what, alongside contrary evidence, so pass the ledger along when you hand the judgement to someone else.
This is general guidance, not legal advice. A scan result is a probabilistic assessment, not proof.