Reading up a bit on the adversarial stuff and apparently there is this image that generally helps fool the different types of neural nets that do IMG recognition. Try putting the png file somewhere in your memes and see if you find less shadowbanning/longer impression time. Not a twitterfag so good luck to those who are on that front.
Dont take my word for it working 100%.
I'd say maybe, I think there's a certain method of makings distortions such that it has minimal impact on readability for humans. There's a lot of high level math theory involved which I am not capable (nor care about) of doing which is why I went looking for a layman's explanation and came across that png image.
It filters the ID in conjunction with hiding all posts that reply to it (but not necessarily filtering the replying ids)