[TOC2]
- [The last /aicg/ thread](https://chub.ai/characters/Anonymous/the-last-aicg-thread-html-edition-cb5609d0)
- [Reverse Moral 4chan](https://chub.ai/characters/Anonymous/reverse-moral-4chan-html-edition-8df042b6)
- [Mongirl 4chan](https://www.chub.ai/characters/Anonymous/mongirl-4chan-html-edition-7a7b9a78/main)
- [/AFOHM/ - Anthro Females on Human Males](https://www.chub.ai/characters/onaholesama/afohm-fd257529) by [onaholesama](https://rentry.org/onaholesama)
- [desuposting simulator](https://www.chub.ai/characters/desu/desuposting-simulator-17296cba) by [desu](https://www.chub.ai/users/desu)
I'll add your card if I see it on Chub or /aicg/.
Each response is composed of 4chan posts. You can enlarge images by clicking on them. With some effort, the AI can even know about the content of the images. A blue board and red board variant is provided. Supported frontends are SillyTavern, Agnai, Risu, and Venus. This works on GPT-4(-Turbo), Claude, and Turbo-16k. Setting max tokens to 1000 is recommended.

I recommend downloading one of the existing sims that seems closest to what you want, going over the greeting, main defs, example dialog and jailbreak, and editing it to make it yours. (When publishing, don't fork, just publish normally.) You'll need to know a few things for that, which I'll cover in the "Essential knowledge" section. I'll cover more details in "Optional nitty-gritty" if you want to know everything about how this works.
In order to pick which card to start from, here's how the existing sims differ:
- **The last /aicg/ thread**: Blue board. All Anonymous. Has an OP. Two types of images: /aicg/-related images (the AI knows their content via their filenames), desu images (the AI doesn't know their meaning).
- **Reverse Moral 4chan**: Red board. All Anonymous. Has an OP. One type of images: illustrations the AI doesn't know their content.
- **Mongirl 4chan**: Red board. Anonymous and namefag posters. Has an OP. One type of images: illustrations and the AI doesn't know their content.
If you don't want to use images at all, remove all the <figure> tags from the greeting, example dialog, and card jailbreak. If you do, continue reading.
How many images to use? I've used over 200 for mine, but you don't have to go that far. 50 should be fine.
You'll have to edit the example dialog, greeting, and card jailbreak to suit the URL scheme you're using. There are multiple strategies:
**Upload a bunch of images to neocities named 1.png, 2.png, etc...**
Then the last line in the jailbreak would look like:
**Upload a bunch of images to neocities with descriptive names like an_alien_fucking_a_human_girl.png, an_alien_laughing.png, etc...**
This is what "The last /aicg/ thread" does in order for the AI to know the content if the images.
Then the last line in the jailbreak would look like:
Yes, this will result in a HUGE {{random}} macro.
**Create waa.ai shortlinks for gelbooru images like https://waa.ai/moralreversal1, https://waa.ai/moralreversal2, etc...**
Similar to uploading them on neocities. I went this route because I didn't want to run the risk of getting my images taken down due to their controversial content. This is what the last line in the jailbreak would look like then:
The first example message MUST be some example of what {{user}} might say.
The second example message must be a plausible {{char}} output, where the first generated post is what {{user}} just said.
The last example message MUST be a {{user}} message containing the text of the greeting's first post, even if you don't intend for that post to look like it was made by {{user}}. This is to encourage the AI to always write what {{user}} sent in its first generated post.
Remember to change the image links to fit your scheme. Do not use {{random}}/{{roll}} in example dialogue, this isn't (universally) supported.
If you edit the dates/times/post numbers, be very careful to keep everything consistent.
The first line MUST be a MARKDOWN invisible text, e.g. . This is because some non-HTML text is required before the tag in the second paragraph, or frontends will discard it.
By default, the first post has the appearance of an OP message, and has a subject. The subject is configured in the CSS variable inside the tag. If you want to make the first post NOT an O, and/or make the first post look like it was posted by {{user}}, then read the "Commented, beautified CSS" section, you just have to remove a few lines.
The CSS must be on ONE LINE for cross-frontend support, and have NO DOUBLE QUOTES inside it. You can safely replace all double quotes with single quotes.
You can use {{random}} and {{roll}} here.
If you edit the dates/times/post numbers, be very careful to keep everything consistent. If you change the OP's post number, you'll have to edit the three occurrences of in the CSS to reflect the post ID that you're using.
If you find stuff that makes the prompt more complicated with no effect like using instead of , or using in some CSS properties, then it's probably for frontend-specific support. I didn't keep track of all of these choices, just don't go trying to trim stuff that might seem useless, chances are it's not.
Every bot message starts with the following template:
This is mostly for Turbo, otherwise it'll forget to make a post out of what {{user}} said, and if you simply ask to have a post ID higher than the last one, the idiot might still fuck it up, saying "the last post number was {second-to-last post number}, so this one must be {same post number as the last one}". I avoid these mistakes by asking to add 100 instead, fuck it.
This was extremely hard to get consistent results for, even on a previous iteration that used a lot more CoT.
If you want to dig into the CSS, here's a commented version.