How to Tell if an Image Was Made With Midjourney
Midjourney makes this hard on purpose, and not for a good reason. Its images carry no C2PA Content Credentials and no known invisible watermark, so there is no machine-readable proof of origin to check. That leaves the image itself: the visual clues and a detector score, both treated as signals. For anything that matters, ask the creator for the original file and the prompt history.
What Midjourney does and does not embed
As of version 8 in mid 2026, Midjourney ships no C2PA manifest and no known pixel watermark. It has been a member of the Content Authenticity Initiative since 2023 but has not shipped a marking implementation. For contrast, OpenAI attaches both Content Credentials and Google SynthID to its generated images, and Google adds SynthID to Gemini and Imagen output. So a present C2PA record actually points away from Midjourney and toward one of the marked generators.
The practical effect is that a Midjourney image and a screenshot of almost anything look the same to a provenance tool. Only the pixels are left.
The visual tells that still work
- The look. Midjourney output is often cinematic, painterly, and perfectly lit, with a shallow depth of field and a cohesive color palette. A casual phone snapshot rarely looks like a film still, and that mismatch is a clue.
- Hands and text. Version 8.1 improved text rendering, so gibberish text is less reliable than it was, but complex hands, jewelry, and small background text still fail.
- Background logic. Follow railings, fences, and window frames. Repeating textures and impossible architecture persist.
- Reflections. Eyes and glass should reflect the same light sources in the same positions.
For the full method, see our guide to spotting AI-generated images.
Use a detector, and know its limits
Score the original file with the free AI image checker or the app, and treat the result as a probability. Midjourney output is polished enough to fall into a mid-range score, and Midjourney’s own aesthetic is a weak signal on its own, so a single number cannot settle it. We explain why in Do AI Image Detectors Actually Work?
Ask for the original
If you are verifying an image for publication or a purchase, the strongest evidence is not a detector. Ask for the original file, the prompt, and the workflow, and compare them with the finished image. A creator who genuinely made the work can supply them. That is the same standard newsrooms use for other kinds of verification, and it does not depend on anyone’s watermark.
Common questions
Does Midjourney add C2PA Content Credentials?
No, as of version 8 in mid 2026. Midjourney has been a Content Authenticity Initiative member since 2023 but ships no C2PA manifest and no known invisible watermark, so its images carry no machine-readable proof of origin.
Does Midjourney watermark its images?
Not invisibly, as of version 8. There is no known durable pixel watermark, so a Midjourney image that has been screenshotted or re-saved looks the same as any other image to a provenance tool.
How do I detect a Midjourney image, then?
You fall back to the image itself: the visual checks for hands, text, reflections, and background logic, plus a detector score. Both are signals, not proof, because Midjourney output has improved a lot at photorealism.
What is the current Midjourney model?
Version 8.2 launched on July 24, 2026, after V8.1 became the default in June. Version 8 improved prompt following and text rendering, which removed some older tells.
Score a suspect image in seconds
AI Detector checks hands-free and never stores your images.
