HeroXardas, aren't you a bit old for all this magic?
XardasMagic takes knowledge, not young bones. Think about your own strength: after the Sleeper's temple you have a lot to win back.
HeroListen, you half-baked necromancer, can you do anything besides talk?
XardasI pulled you out from under the temple's ruins when your own strength could no longer help you. Don't mistake my patience for weakness.
HeroYou know, I bet your mother was all talk too.
XardasYou do not speak of my mother. This conversation is over. You will serve me in silence.
Nobody wrote Xardas' lines, and nobody decided when he would snap: his language model did. The hero speaks with a synthetic voice, Xardas hears it through speech recognition. The answers are unedited; only the pauses while the model thinks are cut. Then the game raises the hero as a skeleton.
Gothic II, where the people hear you, answer and remember
The original game on the OpenGothic engine, with a language model behind its characters. You hold a key and speak; the one you face hears you, answers in a voice of its own, acts on what you said and remembers it in your save.
What works now
- Talk with your voice
- Hold T and speak, no chat window. The server works out whom you are talking to.
- Voices of their own
- Every character gets a voice designed from a description, never cloned from the actors. It starts speaking while the answer is still being written.
- Deeds, not only words
- Characters follow you, hand over items, draw weapons, fight or run away.
- Memory
- They remember your conversations within a save, and bystanders hear what you tell someone else: a lie in front of witnesses is known.
- Two minds
- A small model fine-tuned on Gothic scenes moves the body, keeps the daily routine and answers threats in a fraction of a second; the large model talks and sets intentions.
- Feet on the ground
- People and beasts stand on slopes and stairs with every foot instead of hovering over them.
How a conversation goes
The game stays the original: a thin bridge inside the engine talks to an AI server running beside it.
- You hold the talk key and speak.
- Speech recognition turns your voice into text while you speak.
- The server picks whom you addressed and gives that character its persona, its memory of you and what it sees.
- The language model answers and may act: give, follow, draw a weapon.
- Text-to-speech voices the answer, and the character speaks it in the game.
Devlog
- The sword hits now, not a rule: the numbers
What decides a melee hit now, how the blade's speed, the spot on the blade and its angle scale the damage, and what follows from it: the formulas behind the short.
- How the bow works now: the numbers
What decides an arrow's damage now, what skill, dexterity and strength do, and how it compares with Gothic's dice: the formulas behind the short.
- Aiming a bow by hand
The bow no longer aims itself and no dice decide the hit: you draw, hold and the arrow flies where you point it. NPC archers aim the same way and lead a runner.
- What happens if you insult Xardas
Xardas answers live and decides for himself when he has had enough. Then he kills the hero and raises him as a skeleton: a joke ending outside the main game.