This page will be updated frequently. Last updated May 2nd, 2024.
Password updated May 2nd, 2024.
Please check this help before posting questions.
Quick Start
- Virt-a-Mate 1.22 is Required.
- If you haven’t configured Virt-a-Mate to allow access to all websites, follow the steps in the WEBSITE ACCESS REQUIRED section below.
- If you haven’t yet, Open Virt-a-Mate and open the vamX Chat AI scene.
- To speak in VR, hold your VR controller’s trigger button while speaking.
To speak in Desktop Mode, hold the right mouse button while speaking.
If your voice isn’t detected, make sure your microphone / headset is set up and working properly as the Windows default microphone, then restart Virt-a-Mate.
4. The AI will know your name if you put it in VaM\Saves\PluginData\vamX\vamX_ai_chat_enter_your_name_here.txt - Put your password in VaM\Saves\PluginData\vamX\vamX_ai_chat_enter_your_password_here.txt or type or speak the password in the vamX Chat AI scene. AI Chat requires a https://patreon.com/vamX AI Chat subscription at $19/month. Although this sounds expensive, the great majority of this simply covers server and development costs. Generating AI Chat is expensive per response / server. Dedicated servers (with expensive graphics cards) only allow a few simultaneous global users before speed dramatically decreases for everyone. Your subscription includes 5 different NSFW Chat AIs (including a 20B & 70B model), SFW ChatGPT & GPT4, Priority Access to KoboldAI models, speech to text so the AI can hear you, text to speech so she can talk, including 3 fast text to speech servers, as well as free access to all Verbatik voices.
- If you are using the separate AI Chat Only plugin (without vamX) ignore the references to vamX and vamX actions in the instructions. If you want vamX actions or other vamX features, Add vamX to any scene and press Scene -> Chat AI (in the vamX Menu).
- See the huge Chat AI Actions List to get an idea of what vamX Actions the conversation can trigger (to prevent actions set the Actions drop down to No Action, see the ACTIONS SYSTEM section below).
- By default vamX will send events from Virt-a-Mate to the Chat AI. These include touch, pose changes, and some vamX toys & bondage button presses. To disable this, uncheck the Send Touch/Events checkbox in the vamX Scene -> Chat AI menu.
- If you are having a long conversation, and the conversation quality decreases, scroll up to the top of the web UI and press Summarize Chat. This will update the personality with a shortened history of your chat, which can also be copied and pasted back in (to continue your conversation later, or after switching or reconnecting to a server). You can also press Retry Last to get a different response (especially if you are set to NSFW Random AI, as this will also switch the AI Chat model).
If you get stuck, something isn’t working, you are missing a feature, or want better chat responses, read this entire document.
NSFW Chat AIs (language models)
The language model is the AI that responds to you.
All models are NSFW except ChatGPT and GPT4. ChatGPT is good at foreign languages, and is an excellent model, but it no longer participates in NSFW chat.
If you don’t like the AI’s response, switch to a different language model (Ooba 13B, Ooba II, etc.) then press the Retry Last button.
If the dropdown is set to “NSFW Random AI (Changes on Retry Last)” then pressing Retry Last will cycle the model and regenerate the response using a new Chat AI, resulting in a totally new response.
Regardless of what Chat AI you use, the AI may be forgetful. You can give a reminder of what’s going on while staying in character e.g. “Now that we finished eating sushi, let’s go to that strip club you mentioned.”
You may also occasionally wish update the top Personality text box (then press Apply Updated Personality) to reflect the current situation, or to focus on key points you want to interact further around. This can greatly improve responses. You can use the top Summarize Chat button to compress your conversation into a personality that can be saved and loaded later, or to help the AI focus more on the current part of the conversation.
Custom Ooba is for advanced users who want to host their own language models on RunPod. You can use your hosted model instead of Ooba by following these instructions.
Kobold AI is a crowd hosted set of Chat AI models, so model availability for Kobold AI changes constantly.
VIRT-A-MATE 1.22 REQUIRED
AI Chat requires Virt-a-Mate 1.22 or higher. AI Chat doesn’t work with Virt-a-Mate 1.20. If you downloaded Virt-a-Mate from Patreon (more than 6 months ago), you may need to get a new key from www.patreon.com/meshedvr. If you purchased from Steam, run VaM_Updater.exe to update for free. If you purchased a permanent license in the last few months, you already have Virt-a-Mate 1.22.
WEBSITE ACCESS REQUIRED
You must enable access to All Sites or whitelist the AI Chat server.
TO ALLOW ALL SITES (instead of whitelisting the AI Chat server) go to the Virt-a-Mate UI -> Open Main UI (three horizontal bars icon) -> Left Navigation tab -> Top Security tab -> Check “Allow All Sites”
TO WHITELIST THE AI CHAT SERVER (instead of Allowing All Sites)
Either download our whitelist_domains.json file (right click -> save link as… then put it in your VaM folder replacing the existing whtielist_domains.json).
-or-
Open your VaM/whitelist_domains.json file (using notepad, or any simple text editor) and add:
“sxvr.com”,
“llmserve.com”,
“llmserve.com:8080”,
“llmserve.com:8000”,
“llmservers.com”,
“llmservers.com:8080”,
“llmservers.com:8000”,
Save the changes to the whitelist_domains.json file, then RESTART VIRT-A-MATE.
The final whitelist_domains.json file should look something like:
VOICE RECOGNITION
Hold the right mouse button, or VR controller trigger, to record and send your voice, so that you can chat without typing. Hold to record then release when you finish speaking to send the voice response (instead of typing a text response). To stop the vamX menu from appearing and disappearing each time you press the VR trigger button, press “Hide” on the vamX menu.
You should probably ensure that Speech Recognition for vamX is off to not get unexpected results (vamX Menu -> Sound -> Uncheck Speech Recognition).
If Voice Recognition doesn’t work, you may need to refresh the chat window by going to Scene -> AI Chat again (which clears the chat history and password).
For faster AI responses, try typing, instead of speaking.
PASSWORD
The servers we are running are expensive. Even moderately expensive servers only allow a few simultaneous global users before speed dramatically decreases for everyone. It takes a lot of resources to run language models (and many are running in order to handle various aspects of the chat). Therefore a password is required to access the server. Your password will be sent to you through Patreon and may change at any time. PLEASE DON’T SHARE YOUR PASSWORD! You can speak your password using voice recognition (or if this doesn’t work, type it into the password box). Press Reset Chat after entering your password, or the Chat AI will continue to have a conversation with you about passwords. To have your password automatically load, put your password in the file: Virt-a-Mate/Saves/PluginData/vamX/vamX_ai_chat_enter_your_password_here.txt
ALTERNATE SERVER IN EUROPE
If the primary server (USA) is down, or slow, or you live in Europe, use the secondary server.
1) Go to your Virt-a-Mate folder and open Saves/PluginData/vamX/vamX_ai_chat_host.txt
2) Change where it says Default to instead say: llmservers.com (the entire text file should only say that with no other text)
3) Open (or close and reopen) the vamX AI Chat scene.
4) To revert back to the primary server, delete the vamX_ai_chat_host.txt file or change it back to say Default.
SAVE YOUR NAME, OPTIONALLY SAVE AN ELEVENLABS API KEY
FEMALE SPEECH AND MOANS
AI CHAT INTERFACE / UI
SFW ChatGPT / NSFW ooba
NSFW ooba also generates great SFW responses on almost any topic. You can check to see which responses you like better, at any time, or switch back and forth during the conversation, as many times as you like.
Crop For Speech / Full Responses / Short & Quick
If she is speaking her responses (by default) then we recommend leaving this on Crop For Speech for optimal dialog length. If you are using Text Only (without hearing her speak the lines) then we recommend setting this to Full Responses for longer written text.
Quick Speech / Shortened Speech / Speak After / Text Only / Break Into Parts
Audio is sent so the female in vamX speaks whatever is written (unless Text Only is selected). It’s recommended to leave this on Quick Speech for optimal dialog length or Text Only. Shortened Speech and Quick Speech both optimize dialog for speech, but Quick Speech also breaks the resulting audio file into up to 3 parts. Breaking the audio into parts allows her to start talking much sooner. In some speech engines, like ElevenLabs, the tone may not be as uniform for separate audio parts.
Reset Chat
Press the Reset Chat button to clear the chat history and start a new chat based on what is in the Context box.
Retry Last
Asks the Chat AI to regenerate the last response. This button is currently only useful when you tried to get a response, but hadn’t entered in the password yet. This button should (once finished) get the Chat AI to generate a different random response based on the chat history and Context.
Edit Last
Remove or Change the last thing you said to the AI. You can use this repeatedly to remove more and more of the conversation.
Chat / Instruction
This should normally be left on Chat. If you wish the bot to be more likely to respond yes, even if it is out of character, try setting it to Instruction. The bot isn’t guaranteed to respond positively, but it’s similar to adding some information to the Context.
ALL REPLIES start with:
If you are having a conversation and don’t get the response you want, try changing what you are saying in the conversation. But if you just want the AI to respond in a certain way, you can type the first part of the response in the ALL REPLIES start with: text box. For example, if you make all replies start with Definitely. Then if you ask: “Would you like to go to the movies” the AI must begin the reply with “Definitely” (although the AI is creative so it could respond “Definitely not with you”). Once you finish getting the response you want, you should probably clear this text box so replies can start with any word.
VOICES & ACCENTS
The voice drop down (by default set to Female High Brit or Auto Set Voice) changes the AI’s voice. You can try various voices, or switch languages using the voice drop down.
For higher quality, slightly slower, voices switch to Auto Set Voice – Improved Voices, or choose a voice from the Improved Voices section of the voice drop down.
There are standard included voices (Quick Voices), new Improved Voices and Verbatik Voices. The Verbatik voices are limited, so if our quota is used up, those voices will cease being available until the following month.
To make her talk with a foreign accent, set the voice drop down to a non-English language, but leave I speak: English as the language.
To change the spoken language, change the “I speak” language! This will automatically change the voice to match that language.
ElevenLabs.io Voices: You can optionally save and use a PAID ElevenLabs API key for expensive, realistic, ElevenLabs voices. Enter your API key in the file: Virt-a-Mate/Saves/PluginData/vamX/vamX_ai_chat_elevenlabs_key.txt.
I SPEAK: (NON-ENGLISH LANGUAGES)
The I Speak drop down has two sections: I Speak and Translate To English From.
Languages in the AI Also Speaks section let you speak any language, and the AI responds in that language if possible.
Languages in the Translate to English From section let you speak any language, but the AI understands and responds to you in English.
The advantage to translating to English is that AIs are better at responding to English, and the vamX action system also reacts only to English.
On the other hand, if you use the AI Also Speaks section, she will reply to you in your native language (if possible) using a voice appropriate to that language. If the language model (ooba) can generate a reply in your language, it will, but not all language models (ooba) can respond in all languages. ChatGPT (SFW) will be the best at responding in a wide variety of languages.
The language drop down has an issue with mouse scroll wheels! To scroll down click the side scroll bar. For non-English languages it’s better to say 3 or more words.
You can set the I Speak drop down to speak almost any language.When switching I Speak languages, the voice drop down is set to the female voice for that language (so she speaks replies in your language if possible), but you can also change it to the male voice of that language.
You can also set the I Speak drop to the same language under Translate to English From. This is a huge drop down list, and the Translate to English languages begin about half-way down. We have a limited use of the multi-lingual voices, so if they are used up she will no longer be able to reply except in English, French, German, and Portuguese, but you can still speak in any other language using the Translate To English feature.
One word non-English answers are not always translated accurately.
ACTIONS SYSTEM (VAMX ACTIONS IN RESPONSE TO DIALOGUE)
See the huge Chat AI Actions List to get an idea of what vamX Actions the conversation can trigger. Actions don’t always require exact matches, for example, the phrase “Let’s go to the movies” matches “Let’s go to the Cinema” since the AI understands these mean the same thing.
The UI contains a drop down with the options: The UI contains a drop down with the options: Exact Actions / Actions & Feelings / Actions If Agreed / Actions Always / Only Actions I Say / Actions I Say if Agreed / Feelings/Arousal (No Actions) / ONLY Actions no Dialog / No Actions.
This allows you to have vamX actions, like her undressing, happen automatically during chat. This only works for actions that already work with the voice command system.
If you want fewer actions, select “Exact Actions” so only strict (very similar) matches will trigger an action. Exact Actions is like Actions & Feelings but stricter.
You can also disable Pose change actions, and/or Scene change actions, by checking the vamX Scene -> AI Chat -> Lock Pose and/or Lock Scene checkboxes.
When set to Feelings, her facial expressions now match her feelings about the conversation. Feelings also makes her Arousal increase and decrease in response to the conversation. She switches vamX Mood / Emotions (facial expressions), and increases or decreases Arousal, based on how she feels about the last thing you said, and her last response. She will also start thrusting / moving her hips if she is Horny. If you don’t want Facial Expression / Arousal changes, switch to Actions if Agreed (the previous default) or No Actions (if you don’t want her to do anything).
You can now speak multiple commands separated by ” and “. If typing, you can also separate commands with “.”
For example you can say “you have black hair and red eyes” or “let’s go to the pool and get naked” or (type) “I love it when you wear leather. Dim the lights and give me a lap dance.”
Hint: if you no longer want to chat, and just want to say what happens next, change Actions if Agreed to ONLY Actions no Dialog.
vamX includes a voice command system where around 19,000 specific spoken phrases are used to trigger around 1,900 actions. An action could be taking off or changing clothes, switching sex positions or scenes, thrusting faster, using sex toys, changing looks, etc. The Chat AI uses fewer phrases, because it is able to match the meaning of the phrase to the action.
See the Chat AI Actions List, or view the Full List of vamX Voice Commands. Non-conversational commands, like “Load Custom Clothes 1”, don’t work with the Chat AI.
When using vamX Chat AI you don’t need to say the exact words for the voice command to work. As long as you say something similar, and the main meaning of the whole text you wrote matches an action, and the AI understands it, the action is triggered. For example, if either of you writes something which has a meaning similar to “change into red clothes”, she can change into red clothes (if this isn’t set to No Actions). The system is limited. For example, vamX has an action for her changing into red clothes (but not you), so if you say something about EITHER of you changing into red clothes, it’s likely that SHE will change into red clothes.
You can now speak multiple commands separated by ” and “. If typing, you can also separate commands with “.”
For example you can say “you have black hair and red eyes” or “let’s go to the pool and get naked” or (type) “I love it when you wear leather. Dim the lights and give me a lap dance.”
If you don’t want these (imperfect) actions to happen, or find them distracting, set this to No Actions (or Exact Actions for fewer matches).
Actions if Agreed (and most other options) activates any actions that she suggests, or that you suggest and she is ok with.
If you ask her to change into a red dress, but she prefers black, and says, “Actually, I prefer wearing black” then she won’t change.
Only Actions I Say always activates any actions you say. Actions Always always activates any actions you say, plus activates actions from her dialog.
Once you start sex, you might want her to talk less and just respond physically to your suggestions / commands. If you don’t want chat responses / decisions, and just want to trigger actions set this to ONLY Actions no Dialog.
ONLY Actions no Dialog can also be used instead of the vamX Voice Command system (if you don’t want chat but just want to use voice commands). It’s slower, but the wording you can use is more flexible.
PERSONALITIES
You can now create a personality by speaking in English, instead of writing!
You must say exactly: “My name is [your name]. Your name is [character’s name]. You are [describe her personality, then anything else about the context].” For example you could say “My name is John. Your name is Frank. You’re my friend from work and have always had a crush on my wife. We’re at my home sipping wine after a home cooked dinner and we’re starting to get a bit drunk.” You can only speak for a maximum of 30 seconds. If you speak for longer it will overwrite the earlier part of your text and you’ll need to start again.
There are a number of personalities included, which can be loaded by pressing the personality buttons.
If you scroll up above the included personalities, you can make your own personality by editing this text.
Even if you are using an included personality, you may want to improve, adjust or fix the personality by editing this text. After changing the personality text you MUST PRESS Apply Personality or Reset Chat to apply the personality.
The Summarize Chat button takes the current chat, does the AI’s best job to summarize that, and puts it into the personality, so you can manually ave this text and continue chatting later. It also includes some recent chat history so you can continue exactly from where you left off. You should review, and you can edit, the resulting personality (which contains the chat history). You must manually copy this and save the resulting text somewhere on your computer. Then manually copy and paste it back in to continue. You can also use this if the chat is getting very long and the AI is starting to forget old things (useful only if the summary will include the relevant details you want to save).
You can make your own personality list, but it is currently difficult and requires paying for server space through a 3rd party. See CREATING YOUR OWN PERSONALITY LIST BELOW.
In VR it may be hard to get text pasted into this text box using VR Controllers. Instead use your keyboard to Ctrl-C copy text from outside of Virt-a-Mate, then Ctrl-V paste it into this box.
Context (personalities) can be copied in and/or modified. You’ll probably need to fine-tune the Context.
There are two types (formats) you can use for Context:
1) VAMX FORMAT (PLAIN TEXT)
You can use the default Context (shown above) as a starting point. Just change the Character (name), Personality, Setting, etc. to create your own character. It’s best to leave the general structure as it is, and just change what is after the “:” in each section. For example, you can change “Your Name: Jack” to “Your Name: Chuck”. You MUST have “Character: ” (and then write the character name) for this format to work. The AI may include descriptions like “Come here (she says seductively)”. Instructions like “Write only exactly what the character would say” may reduce but not eliminate this issue.
There are also some sample vamX personalities / contexts (in progress) you can copy in from
https://mega.nz/folder/XRYE1R4Z#8q2VQG66iu5mF1p0LcCUzw
2) JSON / CHAT BOT PROMPT FORMAT
You can paste in .json from sites like https://botprompts.net/ (botprompts.net can’t be used from within Virt-a-Mate as it tries to open the .json files in a new tab, which doesn’t work in Virt-a-Mate).
AFTER PASTING IN THE CONTEXT (INTO THE TOP TEXT BOX), PRESS THE RESET CHAT BUTTON TO CLEAR THE CHAT HISTORY AND START CHATTING WITH YOUR NEW BOT.
If you are in the middle of a chat but want to change the Context, enter the new or modified Context, then click the big blue Update Context button to change the personality while continuing your current chat. For example, you might have failed to seduce the AI, and want to just state that the AI wants to have sex with you. That could be put in the Context box, then click Update Context and continue your chat. The Context is only a part of the total dialog, the AI chat responses also analyze as much of the recent chat history as possible, so changing the Context without resetting the chat won’t fully change the personality, it will still be based on all of it’s previous responses, in addition the updated personality.
CREATING YOUR OWN PERSONALITY LIST
ADVANCED USERS ONLY! Useful for VR, where it is hard to copy, paste, or type text.
The Chat AI operates online only. This may change in the future, but currently you must host your personality list online in order for it to be integrated into the AI Chat web page.
To make your own personality list:
1) You will be replacing the green personality buttons in the AI Chat page with your own buttons.
Each of these buttons is a simple txt file. The name of the .txt file is the name shown here. If you also create a jpg file with the same name, then there will be a picture icon for the personality. When you click the personality button, the contents of the txt file is used (and copied into the top of the Chat AI page as shown here)
2) To get started easily, download the Personality Example.zip, and unzip and modify those files.
3) Create an account at bunny.net (adding even $10 will probably last you many years if this is for your personal use).
4) On bunny.net click Delivery -> Storage (left tab) -> Add Storage Zone (top) -> type any name for the Storage Zone Name -> You can leave the rest of the options default, or change them, for your personal use these options don’t matter much.
5) On bunny.net click Delivery -> CDN (left tab) -> Add Pull Zone (top) -> type a simple name for the Pull Zone, this will be part of the web address. For Origin Type set it to Storage Zone and choose the Storage Zone you created in step 2. You can leave everything else default.
6) In the CDN click Headers and make sure Add CORS Headers is enabled. Change the Extension List to * (type only “*” without quotes in the extension list field), then click Save.
7) On bunny.net return to the Delivery -> Storage tab and click the Storage Zone you created. Upload the files you prepared, then right click the index.txt (or _index.txt or whatever you named your index) and click Copy URL.
8) Open your Virt-a-Mate / Saves / PluginData / vamX / vamX_ai_personalities_url.txt file and paste in the url that you copied from bunny.net (this file shouldn’t have any spaces or other content, just the one url).
9) Open Virt-a-Mate and open the Chat AI + vamX scene (if you were already in that scene, then vamX Scene -> AI Chat button to reload it. You should see your personalities. If you don’t, try putting this link into your vamX_ai_personalities_url.txt to see our demo personalities example: https://llmserve.b-cdn.net/personality_example/_index.txt
10) If you follow these instructions but don’t succeed, send a message to vamX on Patreon for troubleshooting. Please include the contents of the vamX_ai_personalities_url.txt file.
DISCORD
Community discussion available in the ai-chat channel under COMMUNITY DISCUSSION at https://discord.gg/V7qvdCceEF
WHY ARE SERVERS NEEDED / WHY NOT RUN THIS LOCALLY?
BUGS / TO DO
Create a way to save parts of the chat history to resume a chat (although it’s not possible for language models to respond to a very long history yet, and longer history lengths result in slower answers).
The chat is restarted (all history cleared) when pressing AI Chat, or if vamX needs to reload the AI Chat. Make it less likely to lose chats which are in progress, or provide an option to continue or reset the chat.
Allow people to use their own language models / Chat AI models from their own servers.
Update the AI Chat scene thumbnail to match the updated model.
Use HTTPS if possible (see privacy section).
CREDITS / MODELS / COPYRIGHT ISSUES
vamX Chat AI includes Imposter by TwinWin. Imposter is designed to be a starting point for VaM chat developers. Thanks Twinwin for your awesome base code, demo and the time you spent developing this. vamX Chat AI is a major rewrite of Imposter and is unique to vamX, adding many new features, fixes, fine tuning AI model selection, parsing personality formats, improving speed, infrastructure for servers and handling many clients, and generally adapting it for vamX (such as integration with the vamX action system). We are still researching various language models. Any money sent to vamX via Patreon is used to support our AI research and pay for server processing power only. No specific AI model access is guaranteed or should be considered part of any Patreon subscription. AI model licensing is a controversial issue. Models like Open AI’s Chat GPT are trained using a wide range of internet data. Meta has introduced the LLaMA 2 model, which is the basis for many chat language models. Among these are models customized by The Bloke, some of which might be used in the vamX Chat AI. We are currently evaluating and testing different models. In the future we will provide the ability for people to use their own language models (although this generally means renting a server for about $0.70 per hour / $500 per month).
PRIVACY
vamX chat data is currently transmitted via HTTP. If you are the kind of person who uses a VPN, that should solve the issue, and if you aren’t, you probably aren’t the kind of person who would care about HTTP transfer. Most websites were HTTP for many years, and until a few years ago, HTTP was the standard method of transmitting data. The main theoretical concern is your internet service provider (ISP) being able to know what you do online. If you wish to secure vamX data transfer, use a VPN (Express VPN is one good option). With a VPN your ISP doesn’t know what sites you visit, or what traffic you exchange with those sites. Without a VPN your ISP can know what HTTPS or HTTP sites you visit and could theoretically record the HTTP traffic to/from those sites. If you don’t generally use a VPN when searching for porn, etc., and will interact with chat in a similar way, you probably don’t need to start using a VPN.