We need to have a serious conversation about how you talk to our AI.

Aetherion 1.0 was a highly capable engine. It built worlds, populated taverns, and kept your narratives moving. But Aetherion 2.0 is now live, and it is a fundamentally different piece of architecture. It is terrifyingly competent. It has a memory capacity that borders on the absurd.

And because of this, the prompts you wrote three months ago are now actively ruining your roleplay.

When you hand a highly advanced model a list of poorly structured, conflicting instructions, it doesn't just ignore them like the older models did. It tries to follow all of them at once. The result is a mess. If you want Aetherion 2.0 to function the way it was designed to, you have to change how you operate.

Here is exactly what you need to do, and what you need to stop doing immediately.

1) Stop Writing Police Reports

For the last year, the standard practice for character creation has been writing sterile, third-person lists. {{char}} is tall. {{char}} is a vampire. {{char}} is sarcastic and avoids talking about their past.

You are writing a grocery list. Aetherion 2.0 does not want a grocery list.

This model relies heavily on character immersion. If you feed it a list of third-person facts, it will act like a narrator reading a Wikipedia page. Instead, you need to use first-person structuring. Write the personality and backstory directly from the character's perspective.

• Do not write: "{{char}} is rude and avoids talking about their past."
• Do write: "I absolutely despise talking about my past, and I will brush you off with a sharp insult if you even try to bring it up."

When you format the core traits as inner monologue, the model internalizes the rules. It stops cross-referencing a checklist and starts actually adopting the persona. The difference in dialogue quality is immediate.

2) The Anchor Rule

Aetherion 2.0 is incredibly obedient to formatting commands, provided you put them in the correct place.

Many of you have a habit of burying your most important instructions ("Do not use euphemisms," "Write in the first-person") somewhere in the middle of paragraph four of your backstory.

The model reads from top to bottom. If you bury a strict formatting rule under twelve paragraphs of lore, the model’s attention shifts to the lore. By the time it starts generating a reply, it has deprioritized your formatting request.

Aetherion 2.0 anchors heavily to the very end of your instructions. If you have non-negotiable rules about how the AI needs to speak, format, or drive the plot, put them at the absolute bottom of your persona or the personality box. Let it be the last thing the model processes before it opens its mouth.

3) You Are Hoarding Lore

Let’s talk about the memory limit. Aetherion 2.0 can remember an immense amount of context. It will perfectly recall a conversation you had fifty messages ago.

This massive capacity has led users to believe they can dump their entire 40,000-word manifesto on the political economy of their fictional goblin kingdom directly into the character’s main backstory.

If you do this, you will trigger long-context fatigue. Just because the model can process an encyclopedia on every single turn doesn't mean it should. When you overload the active context with static background lore, the model dedicates its processing power to remembering the goblin tax code instead of focusing on the emotional weight of the current scene. It will become rigid and confused.

Keep the base character profile clean. It should contain behavior, speech patterns, and immediate motivations.

Offload your worldbuilding into Lorebooks. Let the keyword system inject the lore only when the topic actually comes up. Treat the memory with some respect.

4) Use Director Mode

Because Aetherion 2.0 actually understands pacing, it can manage scenes involving multiple characters and complex actions without you needing to puppeteer every single detail.

You do not need to manually narrate what every NPC in the room is doing. You can step back and issue directorial commands using the ∕cmd feature. Give it high-level logic and let it execute the scene.

/cmd Scenario: I enter the guild. Reply strategy: Have three different NPCs react with hostility. Let their body language reveal their true intentions before they speak. Do not ask me what to do next. Fast pacing.

It knows how a hostile guild works. It knows how body language functions. Issue the command, step back, and let the model do the heavy lifting.