Known issues first. Everything else goes to WordBlock Labs.

Yes, we know this is one of the biggest AI roleplay complaints. Some people call it head-hopping. Some call it character hijacking. Some call it writing your side for you. We are aware that AI will sometimes write for you or your character. We have resolved this issue to a large extent with our available LoRA. For people who have not or cannot purchase it, we are still working on better prompting and shell-side methods to reduce it further. Right now, our LoRA has fixed roughly 95-98% of this problem in the tested lane.

We are aware that 9B-24B AI models often struggle with quotation marks. They may miss opening quotes, miss closing quotes, or wrap too much narration in quotes. SoloRoleplayer// has not fully resolved this issue yet. It is the number one issue we are currently working to solve. We are experimenting with training methods, prompt methods, and repair methods to improve it. Stay tuned.

We get it. Some of you want to roleplay darker, meaner, stranger, or racier spaces, and many API-based AI providers do not allow that. That is a provider issue, not something SoloRoleplayer// can fully override. If that matters to you, the downloadable 9B Qwen model we chose for the local lane is abliterated and has far fewer, if any, censors. That lane may be more comfortable for you.

Our tested and chosen Qwen model was the better fit for both size and functionality. That is why the Free local lane is shaped around it. We are actively working on larger models for people with more personal compute. Upcoming milestones are 14B and 24B lanes. If that is more your speed, stay tuned. We are testing a few we like, and Mistral Nemo is showing promise.

We have improved this a lot, especially with stronger memory injection and identity reinforcement, but it can still happen. A companion may occasionally slip perspective, drift tense, or lose a visual identity detail like hair, eyes, or what it is wearing. We are actively tightening STASIS, identity refresh, and scene-space continuity to reduce this further.

AI can still sometimes snap back to a room backdrop instead of the smaller live scene space, or forget exactly where bodies and props are. We have improved this with STASIS and active scene anchors, but it is still an area we are smoothing out.

The BYOAI lane is flexible, but API providers all have their own model names, key rules, and safety behavior. If a provider rejects your request, fails a model name, or blocks content, that may be a provider issue instead of a SoloRoleplayer// issue. We are still making that lane cleaner and friendlier. One practical tip: best results usually come from turning up your max tokens to match the RP strength you picked at setup, especially Level 1, 2, or 3 posing style.