Quickstart
- Run aiMultiFool
- Allow the default model to load (~4.8GB, this only
needs to be done once)
- Locate the ‘Card User’ field at the bottom right of
the app and type your name.
- Open ‘File’ menu and select ‘Load SillyTavern PNG
Character Card’
- Select ‘The Construct’
The loaded scene is now a place
where you can create or do anything.
Example: go to ‘People’ menu and
select ‘Chuck Norris’. Challenge him to a fight. You can do anything in The
Construct, even if you have not made an Action Menu item for it yet, just say
something like ‘spawn a kipper’
User Guide
Application Folders
- models-gguf
- Purpose: Store your GGUF format AI models here.
- cards-png
- Purpose: Store SillyTavern PNG Character Cards in
this folder.
- These PNG files
include embedded JSON metadata that define roleplay scenarios.
- cards-json
- Purpose: Store raw JSON Character Cards here.
- JSON files can
also be used to define roleplay scenarios without an associated image.
- chats-json
- Purpose: Optionally save chat sessions to this folder.
- Allows you to
revisit and continue previous conversations.
What Are SillyTavern AI Character Cards?
SillyTavern
AI Character Cards are used to set up immersive roleplay scenarios with AI.
They have become a widely adopted standard for AI prompting.
Formats
- PNG Character
Cards:
- These are image
files that include JSON metadata embedded within them.
- The JSON defines
the character or scenario details for roleplay.
- JSON Character
Cards:
- These are
standalone raw JSON files that describe the roleplay scenario without an
image.
What Are GGUF Models?
GGUF
Models serve as the core “brains” of the AI, enabling natural
language processing and interaction.
Key
Features
- Performance by
Hardware:
- For GPUs without
CUDA cores, stick to models between 3 billion (3b) and 8 billion (8b)
parameters for faster performance.
- Larger models, like
12 billion (12b) parameters, are better suited for GPUs with CUDA cores.
- Recommended
Models:
- Primary Source:
Hugging Face is the best source of GGUF models. https://huggingface.co/
- File Menu
Models:
- Uncensored
12b Model: Ideal for GPU
users, optimized for detailed and immersive roleplay.
- Censored 3b
Model: Lightweight and
works well on CPU-only setups. Suitable for general use.
- Quantization
Recommendation:
- Q4_K_M: Provides an excellent balance between speed and AI
quality, making it a preferred choice for most setups.
Custom Action Menus
aiMultiFool
includes a feature called Action Menus, which allows you to create and
customize your own menus for interacting with the AI. Here’s a breakdown:
- Creating
Menus:
- Use the Options
/ Action Menu Editor to design custom menus.
- Preset menus
(e.g., Card Tools, Character Tools, Scene Tools, Emotions, Acts of God,
etc.) are provided as examples to get you started.
- Customization:
- You can duplicate
these example menus and modify them to suit your needs.
- Arrange the menus
in the tree view by dragging them to reorganize.
- Capabilities:
- The Action Menu
Editor is a robust tool for configuring AI interaction prompts.
- You can create
user-responsive prompts to enhance your experience.
- Resetting Menu
to Factory Defaults:
- If you encounter
issues, you can reset to default settings by deleting the file
aiMultiFool-Action-Menu.json in the application folder and restarting the
app.
- Sharing and
Backups:
- Use the export
feature to share menus with others or back them up for personal use.
Using the Action Menus
- AI Behavior:
- AI might require
prior interaction to respond accurately to prompts in these menus. Engage
in a conversation first to “warm it up.”
- Fine-tuning
prompts may take a few attempts for optimal results.
- Example:
- The menu item
“Commentary / AI Companion Cube” is a great AI commentator for
your roleplay.
Action Menu / Card Tools
This
menu leverages local AI to help you create Character Cards for role-playing
scenarios. This tool offers a completely offline and private way to create
content, avoiding the need for manual JSON writing or external services like
ChatGPT, which may have limitations or privacy concerns.
- Guiding the
AI:
- Interact with the
AI, guiding it as much or as little as needed, to design your desired
character card.
- Buttons will enable
once a valid Character Card JSON is detected:
- Play Card: Play the last displayed JSON card without saving
to disk.
- Save Card: Save the last displayed JSON card to your device
for future use.
- Inject Card: Inject the last displayed JSON card into an
existing PNG image to create a SillyTavern PNG Character Card.
- Troubleshooting:
- If the JSON file
is invalid and buttons are disabled:
- Type prompts
like “complete valid JSON” to fix formatting issues.
- If the AI
forgets the “data” section of the JSON, prompt it by typing
“you forgot the data section.”
TOP TIP: When prompted to enter ‘first_mes’ (first thing you say)
ALWAYS put something in. This ensures you are the protagonist. Example:
‘barman, give me a whiskey’
File Menu
The
File Menu provides options for managing chat sessions, loading character cards,
and downloading AI models. Here’s an explanation of each feature:
Chat
Session Management
- Load Chat
Session:
- Open a previously
saved chat session to continue your conversation.
- Save Chat
Session:
- Save your current
chat session to a file.
- Unlike many AI
applications, this app does not automatically save your sessions. Privacy
is prioritized by default, and saving is manual.
Character
Card Loading
- Load
SillyTavern PNG Character Card:
- Load a PNG
Character Card for roleplay.
- The associated
image will be displayed alongside the card.
- Load
SillyTavern JSON Character Card:
- Load a JSON
Character Card for roleplay.
- If there is a JPG
file with the same name in the cards-json folder, the app will display
the image when the JSON file is loaded.
Model
Loading and Management
- Load New
Model:
- Loads a different
AI model.
- Useful for
switching between models for different tasks or capabilities.
- Download Card
Roleplay Uncensored Model:
- Rocinante-12B-v1.1-Q4_K_M.gguf:
- A powerful,
uncensored NSFW model designed for immersive roleplaying scenarios.
- Works best
with NVIDIA GPUs that support CUDA cores.
- With 12
billion parameters, it may be too large for CPU-only setups.
- Download
General Chat Censored Model:
- Llama-3.2-3B-Instruct-Q4_K_M.gguf:
- A
lightweight, censored general chat model.
- Based on
Meta’s latest Llama 3.2 release.
- Optimized
for resource efficiency, making it suitable for less demanding hardware.
Tools Menu
The
Tools Menu provides features for creating, managing, and inspecting SillyTavern
Character Cards. Here’s a detailed explanation of its options:
1.
Create SillyTavern Card on ChatGPT
- What It Does:
- Opens the
SillyTavern Card creator on ChatGPT. This tool is useful for designing
cards and creating associated images.
- ChatGPT can
generate scenarios in JSON format and create a DALL·E-based PNG image for
the card.
- The JSON data is
embedded into the PNG card for seamless use.
- How to Use It:
o
Design your card and
scenario using ChatGPT’s tools.
o
Download the completed
PNG card and place it in the cards-png folder.
o
Alternatively, if you
prefer to work with raw JSON, copy the JSON data and save it as a .json file in
the cards-json folder.
- Note:
- While ChatGPT’s
SillyTavern tool is powerful, you can also use aiMultiFool’s AI to create
scenarios privately using the Action Menu.
2.
Inject Character Card JSON into PNG image..
- What It Does:
- Combines a PNG
image with a SillyTavern JSON file to create a finished SillyTavern PNG
Character Card.
- How to Use It:
- Useful in
scenarios where:
- You’ve
created JSON data in ChatGPT but lack access to DALL·E for images.
- You’ve found
or designed your own PNG image and want to integrate it with existing JSON
data.
3.
SillyTavern PNG Character Card Editor
- What It Does:
- Allows you to view
or edit metadata directly in a SillyTavern PNG Character Card. Please be
careful not to break JSON formatting.
Other Controls
Options
Menu
- Action Menu
Editor
- The core feature
for creating and editing Action Menus.
- Enables detailed
customization of how the AI interacts with prompts and scenarios.
- Customization
Settings
- Window
Colour: Change the app’s
background colour.
- Font Colour: Adjust the text colour for readability or style
preferences.
- Font Type: Adjust the font for main parts of app.
AI
Interaction Controls
- Stop AI Button
- Interrupts the AI’s
response if it is generating unnecessary or irrelevant content. If you
type new input to AI while it is speaking it will interrupt its reply.
- Continue AI
Button
- Forces the AI to continue
responding. If you press enter in the input box when it’s empty, it will
activate also.
- Rewind Button
- Effectively
rewinds time.
- Custom
Modifier Button
- Opens a panel
allowing setting of states on a scale of 1 to 10. This can be added and
changed in Options / Action Menu Editor
- Restart Card
- Restart the card
you are currently playing.
- New Chat
Button
- Clears the current
session’s context and begins a fresh conversation. It also resets
inference parameters back to defaults.
Scenario
Card Management (JSON Features)
- Play Card
- Activates when a
valid JSON scenario card is detected.
- Allows you to play
the scenario directly without saving to disk.
- Privacy
Feature: Ensures no data
is saved.
- Save Card
- Activates when a
valid JSON scenario card is detected.
- Enables saving a
valid JSON scenario card to your disk for future use.
- Inject Card
- Activates when a
valid JSON scenario card is detected.
- This directly
injects the JSON into a PNG image creating a SillyTavern format PNG card.
This saves manually exporting out the JSON then using Tools \ Inject
Character Card JSON into PNG image.. which is extra steps.
Parameter Settings
(Recommended:
Keep all settings at their default values.)
- General
Advice:
- GpuLayerCount and ContextSize are loaded when the model
starts.
- Other parameters
are reloaded every time text is submitted to the AI.
- Parameter settings
are not saved when you exit the app.
Key
Parameters
- GpuLayerCount:
- Default: 100.
- Loads into VRAM
only if your GPU supports it.
- NVIDIA Users:
- Install CUDA
for GPU acceleration. https://developer.nvidia.com/cuda-downloads
- Example: An RTX
3060 Ti with CUDA is ~6x faster than a Ryzen 9 5900X 12-Core CPU.
- ContextSize:
- Determines how
much of the previous conversation the AI considers.
- Trade-offs:
- Larger
context = More meaningful replies.
- Larger
context = Slower performance over time.
- Temperature:
- Controls the
randomness or creativity of AI responses.
- Recommended
Values:
- 0.8:
Balanced, works well with most models.
- 1.2: High
creativity but may break roleplay scenarios.
- Talkativeness:
Purpose:
This parameter controls how chatty the AI is in a SillyTavern Character Card.
It is an artificial parameter controlled by prompting so we ghost it out by
default.
Scale:
The value ranges from 0.1 to 1.0, where:
- 0.1 represents minimal verbosity (concise responses).
- 1.0 represents maximum verbosity (very detailed
responses).
The value displayed is
read from the SillyTavern Character Card.
You could also modify
the Character Card defaults directly by editing it from the Tools menu.
- Depth:
Purpose:
This parameter controls the complexity and richness of the AI’s responses in a
SillyTavern Character Card. It is a artificial parameter.
·
Depth:
Specifies the level of detail in responses.
Scale:
The depth value ranges from 1 to 5, where:
·
1: Produces simple
and straightforward responses.
·
5: Produces highly
detailed and immersive responses with emotional and contextual depth.
You could also modify
the Character Card defaults directly by editing it from the Tools menu.
Other Parameters:
- Hover over labels
for tooltips or ask the AI for clarification.
- Caution: Straying too far from default settings may lead to
unpredictable results. Setting MinP to 0.02 can produce nice
results.
Known Issues
- GGUF Model
Compatibility:
- Some GGUF models
may cause crashes or issues with prompt formatting.
- AI Behaviour:
- Certain AI
Character Cards may not work as intended, although most function
correctly.
- The AI can
occasionally:
- Ignore
prompts.
- Get confused
by complex instructions.
- Get confused
which character it is player
- Produce
strange end strings at the end of replies.
Development and Support
- Code and
Development:
- Written in C# by
the author with support from ChatGPT 4o
- Support the
Developer:
- Show your
appreciation by buying the author a beer via the link in the app’s
bottom-right corner. Beer makes code good.
- Join the
Community:
- Visit the Discord
Hub for aiMultiFool (link in the app’s bottom-right corner) to connect
with other users, share experiences, and get support.
Disclaimer
The
owners and operators of the aiMultiFool application are not responsible for any
content generated or used within the aiMultiFool application. Users are solely
accountable for ensuring that any content created, shared, or utilized complies
with all applicable laws, regulations, and ethical standards.
aiMultiFool
provides tools for generating and interacting with content but does not
monitor, endorse, or validate the accuracy, legality, or appropriateness of
user-generated material. The app’s functionality relies on artificial
intelligence, which may occasionally produce unintended, incorrect, or
inappropriate results.
We
value your privacy. aiMultiFool does not transmit chat data or user inputs to
external servers. All interactions and inputs are processed in real time, with
no local logs or records maintained. This ensures that all user data remains
completely private and secure, without any possibility of retrieval or storage.
By
using aiMultiFool, you acknowledge and agree that the owners are not liable for
any outcomes, including but not limited to legal issues, damages, or disputes
arising from the use or misuse of the application. Users are encouraged to
exercise discretion and take responsibility for reviewing and using content
responsibly.