Unmodified GAN-generated faces (at least, those in widespread use) have the telltale trait that the primary facial features (particularly the eyes) are rendered in the same location on each image. This becomes obvious when multiple "faces" are blended together.
These "faces" contain additional anomalies that indicate that they are synthetically generated, such as the surreal backgrounds in @Doctor_Iver's and @jonidanna's images and @jonidanna's mismatched earrings. @mikerson_mike's "face" was obtained from generated.photos.
Some (not all) of these accounts push specific political agendas:
@Doctor_Iver: promoted 2022 Canada trucker convoy @Justahappyman: falsely attributes various bad things to COVID vaccines @Kenoisseur: posts crime stats as "evidence" of a (nonexistent) "white genocide" in the US
This is not the first time we've encountered "verified" accounts with GAN-generated faces. Here's a brief thread on the topic from November 2022, when Blue verification first launched:
Over the last 24 hours or so, a bunch of popular accounts that tweet about US politics got followed by a swarm of newly created accounts with default profile pics. What's up with that?
These followers are part of a fake follower network consisting of (at least) 6279 accounts created between January 4th and January 6th, 2023. All have zero tweets and default profile images.
The accounts in this fake follower network all have display names consisting of a first name and last name, sometimes accompanied by one of the following prefixes or suffixes:
DDS
DVM
Dr.
Jr.
MD
Miss
Mr.
Mrs.
Ms.
PhD
Sr.
I
II
III
IV
V
This post covers a series of fake accounts that all appear to be attempts to create the same persona. We ran across the first account (@cheecierom) in July 2020, when it was attempting to spread a rumor about Bill Barr visiting Ghislaine Maxwell in jail:
The second incarnation of this persona - @/romanochee69 (later renamed to @Mollyjucee091) turned up later in 2020, when a photo of a locked mailbox it tweeted went viral:
Meet @SophiaM98566027 (permanent ID 1602043736195751937), a Twitter account created in December 2022 with a stolen profile photo that claims to be a Democrat and 12-year military veteran (allegedly serving in the "Nation Guard").
Unsurprisingly, @SophiaM98566027's profile pic is not the only plagiarized photo posted by this account. The photographs in the account's "good morning" tweets are stolen from the social media pages of random models, and are therefore unlikely to depict the account operator.
This photographic plagiarism habit isn't limited to images of humans. @SophiaM98566027 has repeatedly tweeted stolen cat and dog pics, often claiming or implying that the photos depict @SophiaM98566027's pets.
About three years ago, a bunch of accounts with GAN-generated faces moved their companies to Las Vegas, where their companies were nearly destroyed. #NewYearsShenaniGANs
The duplicate tweets about moving a company to Las Vegas are from a botnet consisting of 37 accounts created on Sept 28th/29th, 2022. All 37 accounts have GAN-generated face pics, and (with the exception of a few early tweets) all tweet exclusively via automation service IFTTT.
GAN-generated face pics (at least, the ones in common usage generated by StyleGAN) have the telltale trait that the major facial features (especially the eyes) are in the same location on every image. This becomes obvious when the images are blended together.
Here's a look at 60 hours' worth of recent growth of one of the largest fake follower networks we've ever seen. Between midnight on December 22nd and noon on December 24th, the network's operators created (at least) 131268 empty Twitter accounts.
We've been tracking various iterations of this massive fake follower network for months, over which the operators of which have created (and lost) millions of fake Twitter accounts. Some previous threads:
The two accounts most frequently followed by these fake followers are @elonmusk and @Twitter, whom the fake accounts followed ~2 days after being created. These fake accounts comprised 23.4% of @elonmusk's and 69.5% of @Twitter's most recent followers as of Dec 26th at 17:30 PST.
It's Christmas Day, so to mark the occasion here's a thread on an NFT spam network tweeting about a "Penguin Christmas" and tagging a bunch of random accounts.
This spam network consists of 830 accounts created on November 21st, 2022. Their @-handles are composed of ten random letters: the first either uppercase or lowercase, the next four lowercase, and the last five uppercase. Most of their display names contain the word "penguin".
Although there are 830 accounts in this network, they only have 23 distinct profile images between them. Each image is used by between 21 and 49 of the accounts in the network.