Two allegations, two evidence paths

EVOX Productions filed against Midjourney in the U.S. District Court for the Central District of California as case 2:26-cv-08408. The complaint alleges that the photographs were downloaded from publicly available online datasets without a licence or consent, and separately that watermarks, image credits and metadata were removed before the files were used for training.[1]

Those two allegations do not stand or fall together. The first turns on how training is characterised, an argument Midjourney has already made elsewhere by calling training on copyrighted material fair use. The second is a claim about what a file contained at one moment and did not contain at the next, which is the kind of assertion a stored copy can settle.[1]

Where the case stands

Nothing here has been found or ruled. What exists is a complaint, and Midjourney has not responded publicly to it. Reading the metadata allegation as established would mean treating a plaintiff's account of the defendant's internal processing as a finding, and the filing does not carry that weight on its own.[1]

The document that would move it is identifiable. If the dataset entries EVOX points to can be produced as they were ingested, the presence or absence of credit fields and embedded metadata is readable from the copies rather than argued from the outputs. A plausible alternative would also become visible in the same place: the stripping may have happened when the public dataset was assembled, before anyone at Midjourney touched it.[1]

Which remedy answers which harm

EVOX says it has photographed nearly every make and model sold in the United States since 2000 and licenses that library commercially, and that subscribers who generate car images stop paying to license its work. It asks the court to stop the use and to order compensation.[1]

Compensation answers the licensing loss. It does not answer the metadata allegation, because the harm there is that a photograph travels without the information identifying who made it. Payment settles an account; restoring credit fields in a dataset would be a different remedy, and the complaint's request as reported does not distinguish between them.[1]