Compositev12

Discovery

Preview on key

You have no API keys yet. Create one on the Keys page.

Use one User: or Assistant: line per message.

Paste asset tags, like the ones Copy tag gives you, to try them on this key without attaching them.

Choose an API key to preview.

Prompts

0 prompts
No prompts yet.

Required

Press Enter or comma to add. Up to 12 tags, 32 characters each; duplicates are removed.

0 characters

Card image
Build with AI

Create an API key on the Keys page to use the assistant. It runs on that key and bills like any request on it.

Lorebooks

0 lorebooks
No lorebooks yet. Import an ST JSON file or create one.

Required

Card image

How many recent messages are searched for keywords. 1 to 50 messages.

The most this book adds to one request. 100 to 8,000 tokens.

Entries

Entries activate when their keywords appear in the conversation.

No entries yet. Add one to start building your lorebook.
Test against a sample chat

One message per line, starting with User: or Assistant:. Entries scan the newest messages up to the scan depth.

No test run yet. Run a test to see which entries activate and how much of the token budget they use.

Build with AI

Create an API key on the Keys page to use the assistant. It runs on that key and bills like any request on it.

Advanced JSON

Supports ST World Info objects (entries) and character-card lorebooks (character_book.entries).

Presets

Markers, regex scripts, and macros beyond {{char}} and {{user}} aren't supported.

0 presets
No presets yet. Import a SillyTavern preset JSON file to start.

Required

Card image
Sections

Enabled sections are sent in this order. System sections join one block at the top; the rest go into the chat at their depth.

Section 1 system block

Shown when choosing which sections a key sends.

Empty sections are skipped when you save.

Sampling and prefill

Applies to the whole preset. Leave a field empty to use the model's default.

Higher is more varied. 0 to 2.

Keeps the likeliest words up to this share. 0 to 1.

Keeps this many likeliest words; 0 is off. 0 to 500.

Discourages words by how often they appeared. -2 to 2.

Discourages any word already used. -2 to 2.

1 is off; higher repeats less. 0 to 2.

Drops words far less likely than the top one. 0 to 1.

The longest a reply can be. 1 to 65,536 tokens.

Test against a sample chat
Build with AI

Create an API key on the Keys page to use the assistant. It runs on that key and bills like any request on it.

A key uses one preset. Unchecked sections are not sent.

Select an API key to configure its preset.

Plugins

0 plugins
No plugins yet.

Required

Card image
ParametersSettings people can change when they add the plugin to a key

Steps run top to bottom on every chat request. A check runs its Then steps when it matches and its Otherwise steps when it doesn't.

When a chat arrives

Runs the workflow on this message without calling a model, and shows what it would add.

Choose an API key to attach this plugin.

Voices

Required

Press Enter or comma to add. Up to 12 tags, 32 characters each; duplicates are removed.

Shown on the public page. Required to publish: say it's your voice, an original voice, or whose voice it is and that you have their permission.

Recording

Use 3 to 15 seconds of one person speaking clearly, with no music behind them. Longer audio can be trimmed below.

A word-for-word transcript helps some models match the voice. 0 of 1,000 characters.

Loading speech models…

Leave empty for the model's normal pace. 0.25 to 4.

Sample

Save the voice first, then make a sample.