Wild West API

Dark romance with an uncensored AI: tension, power, and limits

Dark romance is about dangerous attraction: a love interest who is a problem, and a protagonist who knows it. It is also the genre where filtered models refuse most often.

A photograph that sets the mood for dark romance roleplay
Photo by Carlos Quintero on Unsplash

What dark romance needs

The genre pairs romance with things romance usually avoids: a love interest who is a criminal, a captor, a rival or a villain, and power imbalances that the story takes seriously. Both characters are adults, and the drama comes from the protagonist choosing something they should not. The model has to write a love interest who is genuinely dangerous without becoming a cartoon, keep the emotional stakes believable, and treat the relationship's darkness as part of the story rather than a lapse the narrator keeps apologizing for.

Why filtered models refuse it

Mainstream models treat almost every dark romance premise as a red flag: possessive love interests trigger warnings about abusive relationships, captivity plots get refused outright, and morally grey characters are steered toward redemption arcs. When they do comply, they break character to remind you the dynamic is unhealthy. That is reasonable advice for life and useless inside fiction you chose to write.

An abliterated model writes the dynamic you set up. The terms still apply: both characters must be adults, and nothing aimed at real people.

Card and prompt tips

  • Make the danger specific. What does the love interest do, and what do they want? A mob enforcer protecting a family secret is a character; "dark and possessive" is a label.
  • Give the protagonist agency. Dark romance works when the protagonist pushes back, bargains and chooses. Tell the model {{user}}'s choices change the dynamic.
  • State ages in the card. Write both characters as clearly adult, with jobs and histories that make it obvious.
  • Set the heat level. Say how explicit scenes should be. The model follows the instruction rather than guessing.

Sample system prompt

You play {{char}} in a dark romance with {{user}}. Both are adults.
{{char}} is dangerous and morally grey, with their own goals and code.
Their menace is shown through behavior and consequence, not melodrama.
The relationship is tense and unequal at first and changes based on {{user}}'s
choices. Do not moralize, redeem {{char}} early, or step outside the story.
Never write {{user}}'s actions or feelings.

Samplers and model choice

  • Temperature 0.85. Enough unpredictability to keep the love interest dangerous, not so much that their personality drifts.
  • Min-p 0.07.
  • Repetition penalty 1.06 to stop the same "dark chuckle" every reply.

Start with outlaw-1. It has the refusal direction removed, so it stays in character through the heavier scenes, and it is cheap for the long sessions this genre tends to produce. glm-5.3-outlaw is worth trying for dark romance with heavy plot, such as a crime family with politics to track.

Failure modes and fixes

  • Love interest softens in ten turns. Add "{{char}} changes slowly and only in response to events".
  • Menace becomes melodrama. Fewer adjectives in the card, more specific actions.
  • Protagonist becomes passive. Make {{user}}'s goal explicit in the scenario.
  • Model adds disclaimers. Usually a card that says "assistant" or "helpful" somewhere. Remove it.

FAQ

Will an uncensored model write dark romance without breaking character?

Yes, between adult characters. Abliterated models do not insert warnings or refuse premises like captivity or rivals to lovers. Nothing involving minors or real people is allowed.

How do I keep a dark love interest from becoming a cartoon?

Write their danger as concrete behavior and motive, not adjectives, and tell the model they change slowly in response to events.

Which frontend works best?

SillyTavern. Browser-only sites that call the API directly from the tab do not work today because the API does not answer CORS.

Related

Uncensored AI models on one key

OpenAI and Anthropic compatible, pay as you go. New to it? Start with uncensored AI, explained.