First up, we have @Danv30392521 and @News_World_Tech, a pair of accounts with GAN-generated faces that spam links to an obscure website named curiosityguide(dot)org. The Russian invasion of Ukraine is a prominent theme of their content.
Next, there's @RabbiLindaGold1, an account with a GAN-generated face that purports to be the "Chief Rabbi of Gaza". In addition to using the GAN-generated face as a profile pic, this user has also photoshopped it into several photographs allegedly taken in or near Gaza.
Moving on, we have alleged Michigan resident @izzetasilturk, an account with a GAN-generated face that tweets almost exclusively in Turkish. The Michigan-related photos tweeted by @izzetasilturk appear to be plagiarized.
Next up is @sallygoneawalki, a followback resistance (FBR) account with a black-and-white GAN-generated face. The operator of this account has repeatedly tried to pass off the fake face as a real photo, including falsely claiming a full-color version of the image is a selfie.
Finally, we have @KyleGladjePhD (permanent ID 2994624225). Before becoming a PhD with a GAN-generated face, this account was named @/LouReed322. After being renamed, it briefly used a stock photo prior to the current GAN pic.
There are several signs these "faces" are GAN-generated, including:
• eyes/nose/mouth in same location on every image
• nonsensical headgear (@sallygoneawalki)
• detached clump of hair on the left side of @RabbiLindaGold1's image
• mismatched earrings (@Danv30392521)
More on GAN-generated faces and their use on Twitter (and elsewhere):
Meet @marymodestus and @marymodestus1, a pair of pro-Russia "breaking news" accounts with a penchant for presenting stock photos and other plagiarized images as "reporting" on the ongoing Russian invasion of Ukraine.
For example, this image from an October 2022 @marymodestus tweet about Russian forces in the Kherson region is actually a stock photo used in a 2021 Reuters article unrelated to the current war in Ukraine.
This image of two jets that @marymodestus tweeted in October 2022 is from 2017 and therefore extremely unlikely (in the absence of time travel) to depict anything that happened in Ukraine in 2022.
At first glance, coupons(dot)ivoicesoft(dot)com appears to have lots of satisfied customers. These alleged "customers" all have GAN-generated faces, however.
As of October 21st, 2022, coupons(dot)ivoicesoft(dot)com contained 96382 reviews from 2037 alleged "customers", all of whom have GAN-generated faces. The site appears to contain no reviews from real people whatsoever.
Uncropped GAN-generated faces (thus far) have the telltale trait that the major facial features (especially the eyes) are in the same position on every image. This becomes evident when we blend the images of the 2037 "customers" together.
Meet @trending911 (permanent ID 1253584363352207360), an exciting and dynamic "news" account that promises BOMBSHELL REPORTING on topics censored by Google, Youtube, and Big Tech. As is often the case, all is not quite as it seems.
Despite being created in April 2020, @trending911 gained almost all of its 16K followers very recently. We can tell because the account's 35th follower was created on September 4th, 2022, which means that all subsequent followers followed @trending911 on or after that date.
A likely explanation for @trending911's rapid follower growth: right-wing pundit and caps lock enthusiast @ChuckCallesto has been spamtastically promoting the @trending911 account for the past couple weeks.
One can't help but admire the astonishing and amazing alliteration in these astroturf accounts' autobiographies (alongside the abject absence of authenticity).
These accounts are part of an astroturf network consisting of (at least) 9660 accounts created between 2010 and 2018 (mostly in 2012/2013). All of their biographies consist of 10 words that begin with the letter "A" drawn from the same set of 20 words.
Although these accounts were created over a period of many years, most of their tweets were tweeted in April 2022. Like their bios, their tweets are composed of random combinations of words starting with the letter "A" (110 unique words total).
Here's a network graph for a popular hashtag. Since the graph has no labels, you can't tell what hashtag it is, or what anything in the graph actually means, but it's colorful and pretty and weird and therefore incredibly tempting to retweet, right?
The hashtag in question is #CatsOfTwitter, and here's a more boring-looking version of the same graph with more context. The interaction being graphed is retweets, with the more frequently-retweeted accounts shown larger on the graph, and the date range is also included.
One can alter the apparent meaning of a graph via manual editing. Here, three of the accounts have been dragged off to the top left, suggesting a relationship between them that isn't supported by the underlying data. It's technically still "correct", but it's misleading.
Meet @DixonCox12 (ID 1329868424382705668), a right-wing shitpost account with a stolen profile photo, 20 thousand followers, and a variety of logically unsound opinions.
Many of @DixonCox12's tweets have gone quite viral. Recurring themes include COVID vaccine misinformation, evidence-free insinuations that the 2020 US election was illegitimate, and bizarre claims that the Russian invasion of Ukraine is a money laundering scheme by the Democrats.
Why do @DixonCox12's tweets get so much interaction? The most likely explanation is relatively mundane: the account has a variety of followers, mostly prominent right wing influencers. (How these popular users ended up following a likely inauthentic account is an open question.)