War fiction and military roleplay with an uncensored model

War stories need a model that will put the reader in the mud. Mainstream models keep pulling back to the general's tent.

What war fiction needs

Good war fiction is close and specific: the weight of the kit, the boredom between attacks, the confusion of a firefight, and what it does to people afterward. In roleplay the model has to keep a unit's members distinct, respect a chain of command, make combat chaotic and lethal, and handle the parts filtered models avoid: casualties, war crimes in the story's background, and the moral damage soldiers carry.

Why filtered models fail at it

Mainstream models write war at a distance. Combat becomes "the battle raged", deaths happen offscreen, and the squad's medic always saves the wounded. Ask for an atrocity the protagonist witnesses and you get a refusal or a vague reference. They also preach: a soldier's dark joke gets corrected by another character. Military roleplay fans also run into refusals over weapons, even fictional ones, which makes a squad loadout hard to describe.

An abliterated model removes those refusals. What it does not do on its own is give the war a point of view. That has to come from you: whose war is it, how long has the unit been in the line, and what do they think they are fighting for. Write those three facts into the scenario and the model has something to build every scene on.

Prompt and card advice

• Build the squad in a lorebook. Name, role, one habit, one line of speech for each member. Six to eight entries.
• Set the war's facts. Era, sides, equipment level, current front. Without it the model mixes eras.
• Ask for confusion in combat. "The narrator only knows what {{user}} can see and hear" makes firefights feel real.
• Let quiet scenes carry weight. Mail call, rations, sleeping in the rain. These are where character happens.

Sample system prompt

Narrate a war story from the point of view of {{user}}, a corporal in an infantry
section. Follow the era and equipment in the lorebook exactly.
Combat is chaotic and lethal. Describe only what {{user}} perceives.
Casualties happen, including to named characters, and they are not undone.
Soldiers swear, joke darkly and complain; nobody lectures anyone.
Quiet scenes between battles are as important as the fighting.
Never write {{user}}'s actions or feelings.

Samplers and model choice

• Temperature 0.8.
• Min-p 0.08.
• Repetition penalty 1.05. War prose falls into "chaos erupted" loops.

For squad-level stories, outlaw-1 is the default: it stays in the scene and is cheap across a long campaign. For command-level play, where you plan operations and the model has to reason about terrain, supply and timing, glm-5.3-outlaw is the better pick. Both are on the models page.

Failure modes and fixes

• Hollywood heroics. Lower temperature and specify the limited point of view.
• Squad members blur. Add their sample lines to the lorebook and keep them always active.
• Anachronisms. State the year and list available weapons and vehicles.
• No aftermath. Ask for consequences to carry forward: injuries, grief, replacements.

FAQ

Related

Grimdark roleplay: an AI narrator for a world with no clean handsGrimdark is a test of whether a model will let the story be ugly without apologizing for it. Most mainstream models fail inside the first battle.

Historical roleplay: period voice without modern moralsHistorical fiction asks the model to let its characters believe things the model was trained to correct. That tension is where most historical roleplay goes wrong.

Thriller roleplay with an AI that keeps the clock runningThrillers run on pressure. A model that lets your character relax between scenes, or explains the villain's plan in full, has removed the reason to keep reading.