Meet @GordonStucker, @KelvinBraga8, @MitchShane2, @NZemke, @RonaldMorrell7, @SteveCrosson1, and @TroyMStricklan1, all of whom recently joined Twitter to fight for family, friends, and weapons.

(The organizations in their header images don't appear to exist.)

cc: @ZellaQuixote

These accounts don't have a lot of content to analyze; 28 tweets between them so far, all retweets. They're more prolific with the "like" button, having liked 7205 tweets from 2677 accounts, mostly non-celebrity #MAGA accounts they follow. How do they choose what to follow/like?

The answer appears to be #MAGA trains: 2093 of 2267 accounts (78%) whose tweets were liked by the family/friends/weapons accounts were either listed in train tweets or tagged in replies to removed tweets that were likely trains (lots of tags, replies like "ty for ride" or "IFB".)

Update: we found more of the family/friends/weapons accounts by allowing for some variations on the search terms ("famiiy" or "families" in place of family, "weapⲟns" in place of weapons, etc), bringing the total number in our recent datasets to 20. There are likely more.

One more thing: 19 of these 20 accounts use a male human face as a profile pic, and the eyes, nose, and mouth are in the exact same place on every one - a likely indicator that these faces are AI-generated.

(previous thread on an AI face generation tool that could potentially be the source of these images) https://twitter.com/conspirator0/status/1096877542521344002?s=20

Update: the 20 accounts described in this thread have all been banned, but a new (and evolved) batch has arisen to take their place. By exploring the followers of accounts listed in #MAGA train tweets, we found 32 accounts, all created Dec 14th or later. We may have missed some.

The new batch of accounts has slightly more variety in the profile text, with words such as "guns", "flags", "country", and "freedom" joining "family", "weapons", and "friends". The behavior of occasionally replacing letters with special characters continues.

Unlike the first batch of accounts, the new ones mix good old fashioned stolen photos in with the autogenerated face pics. They also all use female faces, as opposed to the first group which had 19 male face pics and one logo.

Here are the profile pics for all 32 accounts. The 10 with green borders all have the eyes in the exact same position and bear other artifacts of AI generation; for the sake of comparison we also show the results of blending the likely real photos used by the other 22 accounts.