Fallout 4 Installation#

If you get stuck anywhere in the installation process, please see Issues Q&A or reach out on Discord.

Video Tutorial#

For a visual guide to help install Mantella, see the tutorial video from Dead Pixel Gaming below:

Requirements#

Hardware Requirements#

xVASynth#

CPU
xVASynth runs in CPU mode by default, using a single CPU core/thread. Generally, if you can already run Fallout 4, xVASynth should work with most CPU setups.

GPU (Optional)

NVIDIA cards that have CUDA can run xVASynth in GPU mode. Using the same GPU for both xVASynth and Fallout 4 may cause stutter.

Storage Requirements#

This mod requires ~16GB of space when xVASynth and all voice models are installed.

Compatibility#

WARS - Weapon Addition and Replacement Suite conflicts with the Mantella Mod esp.
Mantella requires Windows 10 / 11 (it is yet unconfirmed whether it works on Windows 7)
Quick Trade & What’s Your Name will conflict with Mantella’s alternate activation hotkey (H). You can remove this hotkey activation in Mantella’s holotape [-> Main Settings -> Remove conversation activation option] to resolve the conflict.

Fallout4#

Warning

For Mantella to work properly you have to enable archive invalidation in your Fallout 4 ini file if you haven’t already done so. See here for instructions.

Warning

As Mantella accesses and writes to files within your Fallout 4 folder, it is unlikely to work correctly if you have Fallout 4 stored in Program Files / (x86). Please ensure that you have Fallout 4 stored outside of this folder (eg C:\Games\Steam).

Steam does not allow to create a new Steam Game Library on the same disk. You can either move the whole Steam client outside as described on this Steam Support page or use LostDragonist/steam-library-setup-tool to allow multiple Steam Game Libraries on one disk.

Mantella Files#

The Mantella files can be downloaded from Nexus Mods.

File	Installation
Mantella Software	Extract this folder somewhere convenient to you (if you need some inspiration, you can store it in Documents). Do not store this folder in Program Files / (x86), Desktop, or your Fallout 4 folder.
Mantella Mod	This file can be installed in the same way as other mods with your mod manager.

Text-to-Speech#

Mantella can either be run with xVASynth or XTTS to generate the voicelines for NPCs. It is recommended to start with xVASynth because it is has lower hardware requirements than XTTS, but if you either have a powerful GPU / don’t mind paying to run XTTS externally you can choose this option below.

xVASynth

Download xVASynth via Steam (preferred) or Nexus. Do not store xVASynth in your Fallout 4 game folder.
Download the Fallout 4 voice models. You can either download all models via the xVASynth UI if you have Nexus Premium, or manually via the Nexus mods page:

xVASynth Model Installation Options

💎 Nexus Premium (Quickest)

If you are subscribed to Nexus Premium, (1.) open the xVASynth UI and select “Get More Voices” from the bottom left corner. (2.) Unselect all games except for Fallout 4, (3.) click “Check now” (you will be prompted to login to Nexus), then (4.) “Download All”.

Note

You may need to restart the download a few times for all voice models to properly download.

Warning

If the first voice model never finishes installing. See if the game folder exists within resources/app/models. If it is not there, create it and retry. As this is a known bug of xVASynth: https://github.com/DanRuta/xVA-Synth/issues/59

🌊 Torrent (Slowest, Easiest)

Voice models can be downloaded via a single torrent. Torrents can be downloaded via Bittorent clients such as qBittorent. Note that download speeds vary depending on the time of day. Paste the below magnet link in your browser to receive a popup to open it via your Bittorent client, and set the download location to your_xVASynth_folder/resources/app/models/ without creating an extra subfolder:

magnet:?xt=urn:btih:c9cd4a5f766bee3bdb758ff37578440a9ce2dee2&dn=Fallout_4_XVASynth_Voices&tr=udp%3A%2F%2Ftracker.opentrackr.org%3A1337&tr=wss%3A%2F%2Ftracker.openwebtorrent.com&tr=udp%3A%2F%2Fexplodie.org%3A6969&tr=wss%3A%2F%2Ftracker.btorrent.xyz

🛠️ Manual

If you do not have Nexus Premium, you can also download the voice models directly from Nexus here (under “Optional”, not “Old”). Once you have manually downloaded each voice model into a folder, unzip the models manually into the correct xVASynth folder (xVASynth\resources\app\models\fallout4). Once the extraction is complete, you can delete the zipped voice model files.
The contents of xVASynth\resources\app\models\fallout4 should look like the below:

Note

If downloading all models sounds a bit daunting, you can start with the “Piper”, “MaleEvenToned”, “MaleBoston”, “FemaleBoston” voice models to allow you to talk to most generic human NPCs.
Download the .lip plugin for xVASynth here and download FaceFXWrapper from here (you do not need to download CK64Fixes). Instructions on how to install these are on the .lip plugin Nexus page. Make sure to place FaceFXWrapper in the plugins folder as stated on the .lip plugin page.
When installed correctly, it should look like the below:
(Optional) Download the xVASynth DeepMoji Plugin here. It boosts the emotionality of xVASynth voice models to make them sound less robotic (only available in English). Note that some Fallout 4 XVASynth voice models are older (<3.0) and for those the output will not be improved by use of the DeepMoji plugin.
(Optional) Download the xVASynth Punctuation Pitch Adjuster Plugin here. It will add a high pitch tone at the final syllable for sentences that contain a question mark.

XTTS

Local

Download MrHaurrus’s XTTS API server from here and unzip it.
Download the voice models (latents) folder called “english latent speaker for Fallout 4”. Extract this folder into the same folder as xtts-api-server-mantella.exe above. In other words, you should have a folder called latent_speaker_folder in your XTTS folder.
Download the .lip plugin here and download FaceFXWrapper from here (you do not need to download CK64Fixes). Instructions on how to install these are on the .lip plugin Nexus page. Make sure to place FaceFXWrapper in the plugins folder as stated on the .lip plugin page.
Set xtts_server_folder in MantellaSoftware/config.ini to the path of your new exe, and tts_service to “XTTS”.
(Optional) Configure XTTS settings in the [Speech.Advanced] section of MantellaSoftware/config.ini. Note that if you are using an NVIDIA GPU, the xtts_deepspeed setting can improve response times by 2-4x if you also have xtts_device set to “cuda”.

External (from $0.14/hr)

Make a RunPod account (https://www.runpod.io/) and add some credits ($10 minimum).
Follow this link to open the Mantella XTTS Pod. Choose a GPU to run the Pod with (if in doubt, choose the cheapest). Note that cheaper GPUs are available if you change the “Secure Cloud” setting to “Community Cloud”.
On the following two pages just click “Continue” and “Deploy” (you don’t need to change any settings here).
Once the Pod is running, in your MantellaSoftware/config.ini set tts_service to “XTTS” and xtts_url to https://{pod_id}-8020.proxy.runpod.net/, with {pod-id} being the ID of the running Pod. Eg https://a1b2c3d4qwerty-8020.proxy.runpod.net/.
It takes a few minutes for the Pod to start up, you can check the progress in Logs -> Container Logs from the Pod’s dropdown menu. It will say “Uvicorn running” when it is ready.
When you are finished using Mantella, make sure to delete the Pod so that you are no longer charged! It only takes a few clicks to set up a new Pod so don’t feel guilty about deleting it.

Required Fallout 4 mods#

Warning

Always ensure you are downloading the right version of each mod for your version of Fallout 4. You can check your Fallout 4 version by right-clicking its exe file in your Fallout 4. folder and going to Properties -> Details -> File version. it should be 1.10.163.0 (for Fallout 4 desktop) or 1.2.72.0 (for Fallout 4 VR).

Please follow the installation instructions on each of the linked pages:

Fallout 4 Desktop

Mod	Notes
F4SE	Mandatory: Make sure to get the non VR version. Once installed, run f4se_loader.exe instead of Fallout4.exe.
SUP F4SE	Mandatory: Needed to generate the files needed for Mantella to communicate with Fallout 4
Textinputmenu	Optional: Needed to access text input functions

Fallout 4 VR

Mod	Notes
F4SE	Mandatory: Make sure to get the VR version. Once installed, run f4se_loader.exe instead of the Fallout4.exe.
SUP F4SEVR	Mandatory: Needed to generate the files needed for Mantella to communicate with Fallout 4
Fallout 4 Version Check Patcher	Mandatory: Needed for Fallout 4 VR to accept the mod

Language Models (LLMs)#

LLMs power the creation of responses by NPCs. There are a number of different LLMs to choose from, ranging from free local models to large externally hosted models.

Note

Some smaller models may struggle to handle long term conversations and memory summarising.

By default Mantella is set up to run with Toppy, a free and easy to use model hosted on OpenRouter. It is recommended to start with this model. To learn how to get started, see the OpenRouter section below.

API Models#

OpenRouter (Free Models Available)

Create an account with OpenRouter. Go to the “Keys” tab and generate a new key, saving its value to MantellaSoftware/GPT_SECRET_KEY.txt (do not share this secret key with anyone). This is all you need to do to get started with Toppy, Mantella’s default model.

While Toppy is a good model to get started with, it can fall short when trying to handle complex conversations, in-game lore, or long term memories. You are also limited to 100 requests a day when using free models. To try out a different model, in MantellaSoftware/config.ini set model to a model from the list here (eg undi95/toppy-m-7b). Note that the majority of other models are not free, and you will need to add credits to your account to use them.

OpenAI

Copy your OpenAI secret API key (see here if you need help finding it (you will need to set up an account if you haven’t already)) and paste into MantellaSoftware/GPT_SECRET_KEY.txt. Do not share this key with anyone. Note that you will need to set up your payment details for the API to work. Once completed, in MantellaSoftware/config.ini set the model value to an OpenAI model (eg gpt-4o, gpt-4o-mini).

Local Models#

koboldcpp

Install koboldcpp’s latest release from here. If you want to run koboldcpp on your CPU or otherwise do not have an NVIDIA GPU, download koboldcpp_nocuda.exe under “Assets”. If you have an NVIDIA GPU with CUDA support, download koboldcpp.exe under “Assets”.
Download a local model, such as toppy-m-7b.Q4_K_S.gguf from here.
Run koboldcpp.exe. When presented with the launch window, drag the “Context Size” slider to 4096. Click the “Browse” button next to the “Model” field and select the model you downloaded. Click “Launch” in the bottom right corner.

Optional

Under the “Presets” drop down at the top, choose either Use CLBlast, or Use CuBlas (if using Cuda). You will then see a field for GPU Layers. If you want to use CPU only leave it at 0. If you want to use your GPU, you can experiment with how many “layers” to offload to your GPU based on your system.

Note

Make sure koboldcpp is running when Mantella is running!

text-generation-webui

Install the latest text-generation-webui .zip from here.
Place a local model into the text-generation-webui\models folder (to get started, you can download toppy-m-7b.Q4_K_S.gguf from here).
Paste the text “–extensions openai –auto-launch” (as well as “–cpu” for CPU users) into the installed folder’s CMD_FLAGS.txt file.
Start text-generation-webui and wait for the UI to open in your web browser. Navigate to the “Model” tab, select your model from the drop-down list, and click “Load”.

Note

Make sure text-generation-webui is running when Mantella is running!

LM Studio

Download and install LM Studio from here.
Open LM Studio and download an LLM model.
Go to the Local Server tab, and in the drop-down at the top of the page select your model. If a Windows Firewall message pops up click Allow. Click the Start Server button.
In MantellaSoftware/config.ini, set llm_api = http://localhost:1234/v1/. This is the URL that LM Studio is running your model on. If everything works correctly you should see a message display in Mantella.exe saying “Running Mantella with local language model” when you start it.

Note

Make sure LM Studio is running when Mantella is running!

Other LLM Services#

Mantella has the ability to support other language model services, although these services do need to support outputs in the OpenAI format (like text-generation-webui does via the “–extensions openai” option above). You can connect to any OpenAI-API-compatible service by following these steps:

Paste the OpenAI-API-compatible URL of the service into the LLM Service textbox of the Large Language Model tab in the Mantella UI. These URLs typically end with /v1.
(Only needed for non-local services) Paste the name of a given model of the service into the Model textbox of the Large Language Model tab.
(Only needed for non-local services) Paste a secret key to connect to the service in your GPT_SECRET_KEY.txt file (see the quick start guide to learn where GPT_SECRET_KEY.txt is stored).

Other Known Services:

Service	URL	Models List
Alibaba	https://dashscope-intl.aliyuncs.com/compatible-mode/v1	https://www.alibabacloud.com/help/en/model-studio/getting-started/models
Cerebras	https://api.cerebras.ai/v1	https://inference-docs.cerebras.ai/introduction
Google	https://generativelanguage.googleapis.com/v1beta/openai/	https://ai.google.dev/gemini-api/docs/models/gemini
Groq	https://api.groq.com/openai/v1	https://console.groq.com/docs/models

Whisper#

Whisper via your CPU is handled automatically. Open this section only if you like tinkering

guillaumekln’s Faster-Whisper version of Whisper is used as Speech-To-Text engine by Mantella. The engine is already part of the executable and will download a chosen model automatically when launched. Uses a single CPU core by default when listening to the set default Windows microphone. Alternatively text input can be enabled by setting microphone_enabled = 0 within MantellaSoftware/config.ini.

It is reasonably fast even in CPU mode with the base model. Optionally, to use GPU/CUDA mode, some extra files are required, see Faster Whisper documentation. Note that cuBLAS may already be part of the CUDA Toolkit, so you may only require the cudnn_###_infer64_8.dll files to be beside the Mantella executable. Afterwards enable process_device = cuda under [Microphone] in MantellaSoftware/config.ini.

Setup & Configuration#

Set up the MantellaSoftware/config.ini file with your paths to Fallout (fallout4_folder or fallout4VR_folder), xVASynth (xvasynth_folder) / XTTS (xtts_server_folder), and the Mantella Fallout 4 mod (fallout4_mod_folder or fallout4VR_mod_folder). You will need to set game to Fallout4 or Fallout4VR (be mindful of typing it as per the config.ini options). IMPORTANT: You do not need to set the path for every game or mantella_mod, just the ones you plan to use.
Mod Organizer 2 Users
- fallout4_folder
  By default, this is simply the path to your Fallout 4 folder. If you are using a Wabbajack modlist, you may need to try searching for a folder called overwrite/root or “Stock Game” in your Mod Organizer 2 / Wabbajack installation path and set this as your fallout4_folder or fallout4VR_folder path.
- mod_folder
  You can find the mod folder by right clicking the mod in the Mod Organizer 2 UI and selecting “Open in Explorer”:
Vortex Users
- fallout4_folder
  This is simply the path to your Fallout 4 folder.
- mod_folder
  You need to point mod_folder to your Fallout 4/Data folder. So essentially, you need to take the Fallout 4 folder set in fallout4_folder or fallout4VR_folder and add “\Data” to the end of it. For example:
  mod_folder = C:\Steam\steamapps\common\Fallout 4\Data
Run Mantella.exe in MantellaSoftware. Once it does, it is ready for you to select an NPC in-game via the Mantella gun.

Warning

Do not try to run Mantella.exe through Mod Organizer 2 / Vortex as it will not work correctly.

If you add a new unlisted NPC and for some reason the lip sync isn’t happening in game you may need to close and reopen Fallout 4.

Note

While there have not been issues reported with using Mantella in an existing save so far, please be aware that adding mods mid-game can cause problems.

The Mantella gun and settings holotape will be added to your inventory as soon as the exit from Vault 111 is complete. If you install the mod mid-game they should be dropped in your inventory as well (might require a cell change for the trigger to happen). The items can also be crafted at the Chem Station under UTILITY. You can end a conversation by simply saying / typing “goodbye”. If the NPC responds with “safe travels” then the conversation has ended correctly via this latter method.

Other Tips#

Sound volume#

Sound volume can be set by modifying the value for FO4_NPC_response_volume in MantellaSoftware/config.ini

Hotkey#

As well as via the Mantella gun, conversations can also be started via a hotkey (H).

Text Input#

Text input can be enabled by either setting microphone_enabled = 0 in MantellaSoftware/config.ini. You can respond by text by using the ‘H’ hotkey. This can be changed in holotape settings by entering the DirectX scan code.

Holotape#

Many options can be modified in the holotape [Settings] Mantella Settings, such as NPC actions and radiant conversations.
Fallout 4 Holotape Settings

NPCs not Found#

If the NPC is unavailable to Mantella (most likely a modded NPC which you can add yourself via Adding Modded NPCs), the message “Conversation ended.” should immediately pop up in the top left corner and the conversation will exit. If the only message you see from the starting the conversation is “Conversation ended”, please refer to the “Issues Q&A” section for common solutions to this.

Caching#

Voicelines are cached in the MantellaSoftware/data/voicelines/ folder. If this data takes up too much space over time, the contents of voicelines/ can be periodically deleted.

Issues#

If you are experiencing errors, please see Issues Q&A. Otherwise, please share the details of the errors and your MantellaSoftware/logging.log file on the Mantella Discord #issues channel.

Fallout 4 Installation#

Video Tutorial#

Requirements#

Hardware Requirements#

xVASynth#

Storage Requirements#

Compatibility#

Fallout4#

Mantella Files#

Text-to-Speech#

💎 Nexus Premium (Quickest)

🌊 Torrent (Slowest, Easiest)

🛠️ Manual

Required Fallout 4 mods#

Language Models (LLMs)#

API Models#

Local Models#

Other LLM Services#

Whisper#

Setup & Configuration#

Other Tips#

Sound volume#

Hotkey#

Text Input#

Holotape#

NPCs not Found#

Caching#

Issues#