SamplerLab

I opened the door and saw a ghost. The ghost looked at me. I looked at the ghost

Likely wordghost
Chance53%
Options3
Gibberish0%
words the bot might say next (taller = more likely) past the pink line = banned gold = gibberish, or already said

See a problem
Fix it

Dice
Same number + same settings = same rolls. Change it to re-roll.

The short version

Accurate vs creative

Your bot writes one word at a time. For every word it has a shortlist with odds, then it rolls dice. These settings decide how loaded the dice are. Creativity down: safe and accurate, but boring and loopy. Creativity up: surprising, until it starts picking garbage and you get word salad. The trick is creative enough to be fun, with a filter that throws out the garbage.

Bot problems and fixes7
Same reply every regenerateCreativity is too low, so it takes the top word every time.Creativity up to 0.9 to 1.2
Repeats phrases, loopsIts favorite words keep winning, so it says them again and again.Loop breaker 0.3 to 0.6
Word salad, random wordsCreativity is high and nothing is throwing out the junk.Weird word filter 0.05 to 0.1
Loses the plotA bit too wild. Fun picks pile up until the scene drifts.Creativity down 0.1 to 0.2
Forgets names, odd synonymsRepeat punisher is too high, so it punishes words it needs, like names.Repeat punisher 1.0 to 1.1
Stops mid-sentenceMax length (often called response length) is too short.Raise max length
Rambles, talks for youMin length is forcing it to keep going, or nothing tells it where to stop.Min length 0, stop at your name
Cheat sheet1

Creativity 1.0 + Weird word filter 0.05 + Repeat punisher 1.05. Everything else off.

That's a solid start for most roleplay models. Change one thing at a time. If the model's page lists its own recommended settings, use those first.

Nerd corner: real names and details

The real names you'll see in settings menus. Typical engine order: penalties, then temperature, then the filters, then the dice roll.

Jobtemperaturetop_pmin_ppenalties
Code, JSON, extraction0 to 0.210none
Assistant chat0.6 to 0.80.9 to 0.950none
Fiction and roleplay0.9 to 1.210.05 to 0.1repetition 1.05 to 1.1
Brainstorming1.0 to 1.30.950.05presence 0.3 to 0.6