Part IV · Platform Features · Chapter 21

Models and Settings

Model, max tokens, temperature.

When you're in a chat, you can customise settings through the three-dot menu. These affect how the AI generates responses.

Model selection

The model dropdown lets you choose which LLM powers your character. Some models are better at following instructions precisely. Some are more creative and expressive. Some are faster but less detailed. If your character feels “off,” try a different model. Sometimes switching models is the single biggest improvement.

Max tokens

Controls how long responses can be. The slider goes from 64 to 5,000, with a default of 200.

Range	What You Get	Best For
64 — 150	1 — 3 sentences. Very brief.	Quick chat, snappy dialogue
150 — 300	A short paragraph. Default zone.	Normal conversation, texting
300 — 600	1 — 2 paragraphs with some detail.	Moderate roleplay
600 — 1,500	Full scenes with narration, thoughts, and dialogue.	Rich roleplay, storytelling
1,500 — 5,000	Extended scenes, multiple characters, deep detail.	Epic roleplay, complex scenes

For chat mode, the default (200) works well. For roleplay, you'll probably want 600+.

Temperature

Controls creativity versus consistency. Default is 0.8.

Low (0.0 — 0.4): Very consistent. Can feel repetitive over long conversations.
Medium (0.5 — 0.9): Good balance. Default zone.
High (1.0 — 2.0): Very creative, unpredictable. Use carefully.

If your character feels robotic, try increasing temperature. If they feel random and unfocused, try lowering it.

Advanced: Top P and Top K

Top P (default 1.0): Lower values make output more focused. Reducing to 0.9 can help with consistency.
Top K (default 50): Lower values reduce randomness. Most users can leave this alone.

Tuning order

Start with defaults. If something feels off, try changing the model first, then max tokens, then temperature. Advanced settings are a last resort.

← Previous · Chapter 20Personas Next · Chapter 22 →Social Experience