Just came across a data analysis of Midjourney’s Discord channels—over 16 million records, all scraped from the public general and newbies channels. They claim to have the full data from twenty general channels. As someone who also teaches students how to handle data, I think a few takeaways are worth discussing.
First up, parameter usage. Square 1:1 images make up 66% of the total, and only about 30% bother to specify other aspect ratios. The niji anime model accounts for just around 4%. Variations are pretty popular—about 23% of all generations are variations, and strong variations have become the go-to, meaning people want more distinct results rather than subtle tweaks.
What really caught me off guard is user activity. In a single month, nearly half of all users generated fewer than ten images, and the number of people who were active for just one day almost overlaps with that low-frequency crowd. The author keeps stressing these are “at least” values, since many paid users go to private bot channels to generate, which the public channels can’t capture.
Weekend generation numbers drop noticeably, with heavy users carrying most of the output. This kind of real platform behavior data is way more valuable than those surface-level traffic stats.