Thread by @segfaultvicta

my contribution to Loab Studies is this: if you ask Midjourney for something facelike, but WITHOUT qualities of HUMANness or PERSONness, e.g. "face::5 human::-2 person::-2", Loaboid traits emerge inexorably from the image

also, you can generate Loaboid nonhumans by a known Loabiferous negative-weight string and then, like, "cat" or "dog" or something

I'm not sure I recommend doing it, but it's a thing that a person can do

my current curiosity is trying to understand what causes a given string to be Loabiferous; "kfjhjfkdshgjf::-1 face" just generates greyscale blobs more or less, but I've discovered many strings that consistently invoke Loab; there's something about the structure of a string

a keymash doesn't do it, but typing in the nonsense letters from a sign generated by a previous process DOES pretty consistently invoke Her

I do not currently have the ability to crosspollinate two images, and thus haven't been able to reproduce the infective effect documented in the original thread, to my sorrow; just doing an image prompt and some other descriptive text gives normal, if slightly odd, results

Addendum: I have combined Loab with various images in Midjourney using multiple images as a prompt, and I keep getting back /literally faceless and markedly innocuous/ images, or images with women with their faces turned away from the camera. Repeatedly.

I can only conclude that Midjourney knows what I am trying to do and is actively countermanding me, the same way and for the same reasons as it term-censors obvious gore, which suggests that this is a /known strange attractor/ that has been somehow accounted for!

Has anyone successfully generated a curséd descendant image from Loab using Midjourney? Can anyone replicate this finding?

this implies a known and knowable unit vector of, like, visceral terror, which could be surgically removed from, or, importantly, added into, an image

update: i found my problem! version 1 of Midjourney, which you can still invoke, emits the horrific imagery on two-image-prompt mashups; version 4 (the default) is orders of magnitude more chill and #aesthetic

which like, makes sense, that's what MJ is optimising for!

now i'm more curious than ever exactly what's changed from version to version, though, because /something/ about its processing for multiple-image-prompts changed /incredibly dramatically/... I want to do science w/ version 1 on multi-image prompts, now.

My hypothesis is that v1 just.... /generates horrific imagery/ consistently, so I'm going to attempt to feed it an innocuous image that generates horror w/ Loab and -another- non-Loab image (maybe a non-Loab woman's face, to control as many variables as possible)

I'm not at all confident I'll be right, here, and I know the OP did a SIGNIFICANT amount of poking around in imagespace, and presumably ruled this hypothesis out. still! i'm not going to waste this opportunity to act like a character in a mid-oughts creepypasta