Help! The AI is doing "XX!" (Aka. Understanding Samplers)

Quick Ask Fact Sheet: AI Sampler Settings for Beginners

I want more creativity.

• Increase Temperature (1.2–2.0) to encourage randomness.
• Lower Min P ( 0.01–0.02) to allow more token variety.
• Use Top P (0.8–0.95) to include diverse but plausible options.

I want less schizo words / more logic.

• Lower Temperature (0.7–1.0) for predictability.
• Raise Min P (0.05–0.1) to filter out low-quality tokens.
• Use Top K (20–50) to limit choices to top-ranked tokens.

The model is too repetitive.

• Increase Repetition Penalty (1.5–2.0) for strict avoidance.
• Swap Models or break patterns
• Lower Min P (0.01–0.03) to widen token selection.
• Use Temperature (Min 1.0, Max 1.5–2.0).

The model uses too many strange words.

• Lower Top P (0.7–0.9) to focus on probable tokens.
• Raise Top K (20–50) to include more (but still top) choices.
• Lower Temperature (0.5–1.0) to reduce randomness.
• Raise Min P (0.05–0.1) to filter garbage tokens.
• Use Mirostat (Mode 1, Tau 4.0–6.0) for steady outputs.

Responses are too short or cut off.

• Increase Min Tokens (50–100) to enforce length.
• Adjust Max Tokens (200–500) for longer outputs.
• Disable Length Penalty to avoid forced endings.

---

*For Further Information: GUIDE 1 & GUIDE 2

---