- Description
- Files2
- Images10
- Videos6
- Posts207
- Logs
- Stats
Current section
- Viewing:
About this mod
Mantella has arrived to Fallout4 desktop & VR! Mantella allows you to have actual conversations with NPCs using your voice or text input by using Whisper for speech-to-text, LLMs (ChatGPT, Llama, etc) for text generation, and xVASynth & XTTS for text-to-speech. NPCs have memories of your past conversations and are aware of in-game events.
- Requirements
- Permissions and credits
- Changelogs
See the trailer: See Mantella in action:
Installation How to start conversations
The most up-to-date instructions to download the mod can be found here:art-from-the-machine.github.io/Mantella
How to use the game event key
- During a conversation, press the "Y" key to add to type a text event in the game (e.g. "Piper heard a gunshot"). It's best to use a third person perspective to mesh with the game events that are automatically sent to the NPCs. You can do this multiple times in a row.
- When you actually reply to the NPC, either by speaking in the mic or by typing a reply (default "H" key), all the game events will be sent along the reply to the AI. Note : the game event log will get cleared at the start of a new conversation and can only store up to 5 events at a time. After 5 events newer events will override the the older ones.
Recommended mod(s)
Key Features
- Interact with every NPC, including modded NPCs
- 600+ NPCs have been given their own unique background descriptions
- Support for local, OpenAI, and OpenRouter language models
- XTTS support for voice generation in addition to XVASynth
- Partial compatibilitywith multiple languages. Meaning : all 3.0 XVASynth voice models are compatible with multiple languages but older models aren't.
- NPCs remember past conversations with you
- NPCs are aware of in-game info such as their location, the current time, picking up and equipping items, sleeping, taking regular or radiation damage, entering power armors, being given commands or settlement assignments.
- NPCs can start conversations with each other (Radiant Dialogue)
- Can have a conversation with two NPCs at once (don't recommend going above that for now)
- Fully playable in desktop or VR (make sure to download the appropriate mod)
- (experimental) NPCs can be convinced to follow your commands, or may attack you if you offend them
FAQ
Can I use Mantella with text input?
Absolutely. You have to disable to disable the microphone in the config.ini (see installation docs, link above) then press "H" in-game to respond to NPCs (desktop only).
Can I use Mantella with VR or desktop?
Yes it's compatible with both. Make sure to download the right mod.
How much does it cost to run Mantella?
At a minimum, nothing. If your hardware is strong enough to support running local language models you can run completely offline for free. If you would prefer to outsource this processing externally, you can also use OpenAI or OpenRouter:
OpenAI: The OpenAI API (note: not ChatGPT Plus!) is billed on a monthly basis (with the first $5 free) on a pay-per-use basis. So far the most I have spent in a single month of (part-time) development is $3.40. Some early adopters have reported up to $7 a month for heavy usage.
OpenRouter: The OpenRouter API is billed by using pre-paid credits (with the first $1 free). There are a number of language model to choose from, and there are often models available for free. See here for more details:https://openrouter.ai/docs#models
Are there high PC hardware requirements to run this?
External service minimum specs:
If your are using external LLM services like ChatGPT or Openrouter then the only requirements are running XVASynth and Whisper (if using Speech-to-text for microphone). XVASynth and Whisper can be run easily on most modern CPUs. My video showcases uses a AMD Ryzen 7 3700x with the external LLMs and I get between 2-4 seconds for response times.
Local minimum specs:
A local LLM will need a minimum 6 gb of free RAM & VRAM to run a basic role-playing capable LLM.
Local minimum specs:
XTTS Specs: Using XTTS will add another requirement of 2-4 gbs of RAM and VRAM.
What are the storage requirements for this mod?
XVASynth without models needs about 4 gb. XVASynth requires ~15GB of space when all xVASynth voice models are
installed (and 10 gb for temporary files that can be deleted after installing the voices). MantellaSoftware itself is 200 mbs but will need 1-5 gb of temporary files.
Does this work with local language models?
Yes.
Compatibility
- WARS - Weapon Addition and Replacement Suite has been reported as a compatibility issue.
- Fallout Mantella is compatible with Windows 10 / 11 (tested for both)
- Quick Trade & What's Your Name will conflict with Mantella for the alternate activation option. You can use the holotape followed by [-> Main Settings -> Remove conversation activation option] to resolve the conflict.
Credits
Supported Languages
TommInfinite for SUP F4SE & SUP F4SE VR
Neanka for textinputmenu
kinggath for parts of the ListenerScript
Developers
- Art from the Machine (thank you Art for Mantella itself and all of your support while making this Fallout 4 mod)
- Pendrokar
- Leidtier
- Mahe
- mercury7
- MrHaurrus
- Pathos
- teddybear082
- tonycubed2
Issues
Uninstallation
please keep your MantellaSoftware/data/conversations folder somewhere
safe)
Community
If you would like to stay up to date on the project or would like share your experiences (and issues!) please find the link to the Discord here:https://discord.gg/Q4BJAdtGUE
Swing by my youtube channel for Mantella video showcases.