Mutt evasion logs
As of November 11th, 2023, trannyjannies implemented a spam filter confirmed by Froot which auto-deletes the original impish mutt image. It seems to work on an AI detection system, the exact workings of which are unknown. Many 'teens have developed ways of circumventing this system, but there is a persistent problem: Most of these methods end up compromising the image's composition and readability in very obstructive ways. If the image is too abstracted, too hidden, or too different from the original mutt image, it's essentially not the same image anymore, meaning the jannies win. Thus, a new solution was needed.
In response, a team of dedicated soyentists have begun an effort which promises salvation for obsessed faggots everywhere: To create the closest possible analogue to the original impish mutt image. This is the Great Work of Soyentifical Alchemy, or the Final Transmuttation.
This will be done via a rigid adherence to the fundamental principles of Soyence: Observation, hypothesis, trial-and error testing, and further hypothesizing from results. Enumerated below is the method generally considered the most effective:
-Based on observations on what about other's posts gets past the filters, come up with an idea for an effect to apply to the OG mutt for it to pass the filter, and edit it accordingly.
-Note that you should start out with the version of your idea which is least obscured, and least likely to work. That way later you can obscure it more incrementally with each post to see where the cutoff point is where it stops detecting it. This is the best method because your failed attempts will be deleted instantly, not clogging up the log and drawing less attention to your experimentation.
-Post the mutt, making sure to type out an effortful muttpost to go along with it so mods have less of a justification to delete it for spam reasons.
-Wait for roughly 20 seconds. If your post isn't deleted by then, it's safe to assume that it wasn't detected by the bot.
-If it was deleted within that time frame, and it's clear that jannies are online and deleting posts, make sure that it wasn't deleted by a janny by posting another mutt that you already know isn't detected by the filter, preferably another obscured version of the OG mutt. If the effective one is still deleted, you can have a reasonable suspicion that your muttpost was deleted by a janny, not the bot.
-If you are almost certain that your mutt was deleted by the bot, make a small but still noticeable change to make it less recognizable according to whatever method you've chosen. Make your method more extreme incrementally.
-Every time you post it again, add another 'or whatever' or 'or something' to the text you wrote so that the flood detection doesn't go off.
-Repeat this process until it successfully evades detection.
-Once it successfully evades detection, try to track back and make it more recognizable, either by employing a new technique using what you've learned about how the bot works, softening your previous technique.
Feel free to add a test log below, building upon the research of those before you:
TEST LOG ONE
This experiment session was very fruitful. It yielded an image which almost possesses all 7 of the Amerimutt's features accurately: head, stubble, ears, eyes, nose, mouth, and shirt.
Firstly, I found out that it doesn't detect the body of the mutt without its facial features, or the disembodied facial features of the mutt piled up in a disorganized way.
Judging from this, I could split into two efforts: trying to build back the body from the face, and trying to build back the face onto the body.
I tried the mouth by itself, which it didn't detect.
Then I tried putting an ear and the eyes+nose disorganized on the body, which it didn't detect.
Then I tried the body plus the mouth, which it did detect.
Judging from this, I could assume that the body was an important factor in it detecting the face.
I tried redrawing the ears, which didn't work.
I tried the same face with a new body drawn, which it did detect.
Then i tried doing the face with skin but no body, which it didn't detect
and then i tried doing the whole image completely redrawn but lazily, which it still detected
Judging from this, I could safely assume that it wasn't strictly detecting the individual groups of pixels which made up each part of the face, but rather holistically determining the content of the image.
To see how far its pattern recognition went, i tried putting very minimal facial features (very bare bones, stickman tier) on the new body, which it didnt detect
This told me that it didn't have a function to detect these features based on what they were from the shapes, their color or position (eyes, nose, mouth), only in relation to how much they resembled the original parts of the image.
and then started filling in the respective facial features from the OG mutt image onto the stickman facial features one by one, starting with the eyes and nose
Surprisingly, for both the eyes and the nose, this worked without a hitch.
However, when I tried adding the mouth, it detected it.
I tried redrawing the mouth fairly detailed, but that didn't work either.
In fact, puzzlingly, it seemed that even if I added a mouth that was just a few black lines, it would get detected.
I was then trying to gradually make the stickman mouth more detailed while avoiding using lines to see at what point it got detected, but jannies finally woke up and banned me for spam.
Below is the final image from this session, plus another that is confirmed to pass detection:
