Happy Halloween, folks!
Wanted to share something in season to mark the occasion.
The Anima version is split into two different models. So, the following guide and explanation applies ONLY to
The model's Illustrious version only has two subsets:
1) The usual one you are probably familiar with - green theme, vignetting, through peephole view. This should be used, like the original images, in landscape orientation, but works in portrait or square just as fine.
Trigger is "pknohg", but to make the model give you the peephole view also add "screencap background, vignetting, monochrome, green theme"
2) The color, pastel, pencil like art style that is in the game. Personally, I like it too and wanted to have a single model to make both it and the "meme".
This one is triggered by "pknohc".
*) Both trigger words and substyles can benefit from "horror \(theme\)", so feel free to add that.
Also, the model has "dialogue box" trained in, but obviously when it produces it the text is gibberish. Another helpful tag could be "fisheye lens". Keep in mind that, especially in landscape orientation you might often get more than one individual. That is because of the dataset (FEMA annoying as always). To counter that, use the tag "solo" with added weight - example: (solo:1.3).
Now, finally, I leave you with several questions: Are you a Visitor? ... Am I? ... Are you alone? ... knock knock
...
If you want, you can join my newly made Discord server https://discord.gg/QWx6HzzyYm where I plan to add announcements of upcoming model releases, dedicated section for requests, and where we can hang out and discuss AI, how terrible of a people we are for liking and using it, and for the ungodly and disgusting purpose of P/C0rn on top of that. But also mull over other creative pursuits like music, writing, and generally lots of philosophizing.
Description
This version was trained on the At the Door / Through spyhole looking outside / green theme images: both from the original game and some fan artwork. The added fan art heled with expanding expressions, positions and body proportions, but has diluted the textured, rough look of the style.
This version was trained on mixed captioning and with the adaln module.
Dims are 32, and alpha is 32.
Now, thanks to Anima's great textual understanding, the model can grasp test and text boxes and where to put which words fairly accurately. Use prose (natural language) to guide it.