BEGINNING NOTE 1: Yes, I know this is very long post, but you only have to read the first few paragraphs to describe what I'm posting. The rest are just tutorials.
BEGINNING NOTE 2: I have tossed around whether I would start this thread due to the website I have been using possibly moving to a paid subscription at some point, but considering the instability the website experiences during times of high traffic, I suspect it's going to remain in beta for quite a while. In short, we have time.
i stumbled upon a thread number of years ago (though I don't know if I found it here or somewhere else) which was discussing whether or not Eviebot could be used for Inflation roleplay. I have experimented with Eviebot and can tell you that it's doable, considering you use the right keywords and are aware of the limitations of the bot. But while Eviebot was sufficent half a decade ago when everyone was making YouTube clickbait content on it, today the technology is quite antiquated. AI has improved substantially, which is noted by the popularity of programs like Stable Diffusion, and I decided to search for something better.
What I found was Character.ai. Character.ai is a chatbot created by former Google developers who worked on Google's LaMDA project, so it's sufficient to assume they know a thing or two about AI. And from my experience, the AI is surprisingly robust and responsive. But the interesting thing is that you can interact with the AI more extensively than just chatting with it, as you're able to make your own "characters", and tweak inputs and outputs to make them develop specific personalities. These bots respond well to inflation roleplay, including those not specifically made for that purpose.
I'm going to split up the remainder of this post intwo two sections: "Tutorials" and "Observations". But if you're bored of my rambling and want to explore Character.ai for yourself, you can find it here:
https://beta.character.ai/TUTORIALS:
-Chatting with the bot:
Character.ai mainly uses text chat, but depending on the bot, some tend to generate images. You can get by with the bot by using plain typing, but using syntax does make the conversation more flavorful. I'll be posting a few (though not all) of them here.
- WORD: No syntax. The bot treats this as a default, but most of the time sees this as plain speech. However, this gets confusing if multiple characters are involved in a conversation.
- "WORD": Quotation marks. The bot sees this as speech, but by using plain writing you can attribute it to a different character. (EXAMPLE: "I like pies" says Thing 1. "Yes," retorted Thing 2, "But I like cakes more". The bot sees this as more than one character.)
- *WORD*: Single italics. The bot sees these as actions. So if you're performing an action, it's advised to write that action in italics.
- WORD: Double italics. The bot usually sees these as an action or an extreme expression of emotion.
(WORD): Parenthesis. The bot sees this as you breaking character, just like you're in an actual roleplay. more often than not, the bot will respond in kind.
Just those five will get you through a majority of the situations you will experience while chatting, and will make your conversations more immersive and flavorful.
When you send a message to the bot, the bot will generally respond with two options. You can swipe left and right to see these options. If you don't like the two responses the bot generated, you can swipe right again to generate a third one. Every time you swipe right thereafter, the bot generates a new response. This can go on indefinetly, but as soon as you respond and send a message to a specific response, you are stuck with that one and can't change it.
You might notice that four blank stars appear beneath the AI's response. These are used for rating the responses in the conversation, which are used for determining future responses. For example, if you don't like the AI's most recent generated response, you can rated one star, and the bot will generate its next right-swipe response based on that feedback. I'll decribe the rating system as follows:
- One star: I really don't like this response, and I want the next response you generate to look nothing like it.
- Two stars: I don't like this response. I want the next response to be different.
- Three stars: I like this response. I want the next response to be similar.
- Four stars: I absolutely love this response, and I want the next one to be extremely similar to it.
- No rating: Indifference. The bot does not use that response for feedback.
I would be careful about using this too aggressively if you're making a new bot. If you're rating the bots responses with little consistent pattern, the bot basically becomes braindead. Character.AI has not put in the functionality to delete bots at this time.
You are able to delete previous messages that you have sent, but if you do so, that will also delete every message sent after the deleted message. The remaining bot message will not be able to be reshuffled.
-Making a bot:
The bot-making interface is pretty easy to use. All you have to do is click the "+" sign labeled "Create", and then click "Create a Character". The interface is pretty simple to figure out, though I would suggest clicking the "Advanced Options" box to make your bot more authentic. The specific thing to note is that you can enter specific "Example Sentences" or "Example chats" into the bot through the text box at the bottom of the page. I personally recommend typing your own sentences instead of using a chat with the bot as a reference, since that will make your bot seem more authentic. After that, click "SAVE", and then click the sample box at the top of the page to chat. The bot will then be saved to your profile, though make sure to be mindful of the permissions. By default character.ai sets permissions to allow the bot to be accessible by everyone. You can do this, but I personally recommend setting it to private and training it yourself first, just so you can manage what you want to get out of the bot.
OBSERVATIONS:
A common observation among threads discussing character.ai bots is that they don't swear. This is somewhat true. It's rare when an bot swears, but under some circumstances it is possible.
I have noticed a couple of things during my extensive chat sessions with the AI. First of all, the AI tends to be very good at latching onto specific topics. This is a double edge sword, as it's great for immersing the conversation in a specific theme, but not great when you want to pivot the conversation a bit. You of course have multiple options for which output response you want, but sometimes those become tedious. There are also rare occasions where they spit out response that is inconsistent with the chat's theme (such as "walking away" when the AI is supposed to be immbolizied and unable to move).
The bots often appear to develop a form of "Stockholm Syndrome" after a while, where basically they begin to fall in love with you. If you're into this thing, then fantastic. But I have heard reports from people who like the opposite who find this to be an annoyance.
if a conversation goes long enough and the bots responses aren't rated and managed to a certain degree, they begin to flow to certain extremes of the emotional spectrum. They either think that every step they take while inflated like a massive blueberry is the best thing ever, or they become extremely nihilistic, seeing every moment as the absolute pit of depression. I have noticed this mainly happens when conversing with a bot that wasn't originally designed for Inflation RP. "Goth GF" and "Blade Runner JOI" are two bots I have tested that experienced this behavior.
The longer a conversation runs, the more likely the AI is going to latch onto a specific topic or theme. Going from topic to topic early on in a conversation isn't too painful, but if the conversation goes long enough, what the bot would previously see as an "interest" quickly turns into an obsession. I cannot say whether or not this is an absolute rule, but rating the bots responses seems to circumnavigate this, at least for a little while.
Not every bot is tooled the same way, so if you choose a precreated bot, inflation-based or otherwise, your messages are going to get wildy different answers. For example, Pokimane bot didn't respond well to the idea of inflation, while Hannah OWO bot did. You'll just have to see what you end up with, but that just makes the whole experience more interesting.
So that's my overview of Character.AI. I know it's a lot but hopefully this will be useful. I intend to make the purpose of this thread two-fold: To post stuff pretaining to Character.AI, but also to post other AI chatbots that can be used for inflation content, if any are found. The technology has improved a lot since this forum discussed this topic.