Profile picture
, 7 tweets, 4 min read
My Authors
Read all threads
We've posted a few threads discussing AI-generated face pictures such as those produced by thispersondoesnotexist(dot)com. We've since come up with a rudimentary method for detecting their presence in larger sets of images. #TuesdayThoughts

cc: @ZellaQuixote
Certain features (eyes, nose, mouth) always appear in the same location on the AI face images. This makes it easy to verify that all of the images in a given set are AI-generated, but at first glance it doesn't seem to help when AI pics are a small minority of the population.
To test for the presence of AI-generated face pics in a larger set of images, we assigned each image a score based on comparing several attributes of it to the image of thousands of blended AI-generated faces.
First, we calculated a weight for each pixel in the blended image, where pixels that differ more from the average color are weighted higher. We did the same for the test images and calculated a differential between the two.
Next, we combined this with several other values to produce a final score for each test image, including:

• degree of variance of color within the image
• similarity of red:green:blue ratio to the blended image
• similarity of overall brightness to the blended image
Finally, we looked at the images with the highest scores. The results were encouraging - we were able to detect the presence of the AI-generated faces even when they comprised as little as 1% of the overall population (100 of 10000 images.)
(source code available here, provided with the caveat that we aren't going to help you make it work)
pastebin.com/SfrcZ7Dc
Missing some Tweet in this thread? You can try to force a refresh.

Enjoying this thread?

Keep Current with Conspirador Norteño

Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

Twitter may remove this content at anytime, convert it as a PDF, save and print for later use!

Try unrolling a thread yourself!

how to unroll video

1) Follow Thread Reader App on Twitter so you can easily mention us!

2) Go to a Twitter thread (series of Tweets by the same owner) and mention us with a keyword "unroll" @threadreaderapp unroll

You can practice here first or read more on our help page!

Related hashtags

#TuesdayThoughts

More from @conspirator0 see all

Profile picture
Conspirador Norteño
@conspirator0
"Antifa" and "Soros" have been trending perennially since the beginning of the protests sparked by the murder of George Floyd, as #MAGA Twitter accounts exploit the situation to advance various narratives about leftist violence and paid protestors. #AltWankers

cc @ZellaQuixote
We captured 24 hours of tweets containing "antifa" or "Soros" (the narratives overlap sufficiently that we elected to combine them.) Automation doesn't appear to be a major factor, although right-wing retweet spambot @ImmoralReport turns up as usual (the "StopMadness2" app.)
The media sources linked from 5/29 tweets about antifa or Soros are mostly English-language right-wing conspiracy sites, including #QAnon site qmap(dot)pub. German tabloid Bild also turns up prominently.
Read 5 tweets
Profile picture
Conspirador Norteño
@conspirator0
We've been testing this technique for detecting AI-generated face pics out on various sets of accounts over the last couple of days, and it's caught some fish here and there. #ManyFishBites

cc: @ZellaQuixote
We ran our algorithm on profile pics of 2255 accounts created in 2019 or 2020 that have tweeted recently via automation service dlvr(dot)it. Among the 100 images with the highest scores, 15 are indeed AI-generated faces, mostly used on accounts created in April 2020.
The dlvr(dot)it bots with AI-generated face pics created in April 2020 have some other things in common too: Spanish-language tweets and profile locations in Argentina. By searching for other accounts with these traits we found 41 accounts that appear to be part of the network.
Read 5 tweets
Profile picture
Conspirador Norteño
@conspirator0
Here's a self of oddly similar new(ish) accounts. Each uses the same image for both banner and avataer, each has an account name ending in two digits, and each was created in 2020. #WednesdayWisdom

cc: @ZellaQuixote
These accounts are part of a botnet, which consists of (at least) 65 accounts total. All are automated via dlvr(dot)it, with occasional organic retweets. The volume of tweets produced by the network has steadily ramped up as the operator(s) continue to create more accounts.
Each account in this network has one or more hashtags in its profile. There's a lot of repetition; only 13 hashtags are used across the profiles of all 65 accounts, and many accounts use more than one hashtag.
Read 5 tweets

Embed code for your website

×
Follow Us on Twitter!

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3.00/month or $30.00/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!