Preset Basics

Presets control how the model writes. They combine generation parameters and prompt structure so you can reuse a style across chats.

Who this is for

  • Users whose replies feel repetitive, too short, or too chaotic.
  • Users who want to understand Temperature, Top P, and Min P.
  • Users who need different settings for different characters or models.

What you will learn

  • What a preset is.
  • How common sampling parameters affect roleplay.
  • How to manage global presets.
  • How to override presets for a specific chat.

Key sampling parameters

1. Temperature

Temperature controls randomness. Higher values can make replies more creative, but too high can cause unstable or nonsensical output.

2. Min P

Min P filters low-probability tokens relative to the most likely token. It can help preserve creativity while reducing incoherent output.

3. Top P and Top K

Top P and Top K limit the candidate token pool. If Min P is active, you may keep Top P less restrictive unless your model benefits from tighter filtering.

4. Max Response Tokens

This controls how long a single response can be. Increase it for longer prose or detailed scene writing.


Manage and switch presets

Tavern Studio preset list
The preset list keeps reusable generation configurations organized.

1. Global preset management

Tavern Studio preset entry editor
Preset entries let you tune prompt structure and generation parameters.

Open Presets to create, duplicate, import, and edit preset configurations.

2. Chat-level preset overrides

Open Chat Settings to assign a specific preset to a character or conversation when it should behave differently from the global default.


FAQ

Why did higher Temperature make the reply worse?

The value may be too high for the model. Lower it and consider using Min P to reduce unstable tokens.

Why did my preset changes not affect the current chat?

The chat may have a preset override. Check Chat Settings for that conversation.

Why does the model repeat itself?

Try adjusting repetition-related settings, response length, and prompt structure. Some models also repeat more than others.


Next steps

Download Tavern Studio
Windows