By skyforbes I am running an AI girlfriend service, and want to improve the whole roleplay engine. My main issue currently is that sometimes, the AI details, and produces gibberish characters or long nonsensical sentences. I have already experimented with temperature, frequency penalty, and presence penalty, but the issue still persists for some users.
Currently using Mistral Large 2411 via OpenRouter.
I am looking for someone with experience in:
- LLM selection/tuning for adult conversational realism
- NSFW AI roleplay systems
- Prompt/behavior engineering for immersive characters
If you have real experience with this domain, please DM me.