IMMRAMA is built on a foundation of human creativity. Every word of the game's narrative was written by a person, and every piece of art was reviewed and approved by hand before it shipped. The illustrations were produced by an artist working in Adobe Creative Cloud with an AI-assisted pipeline, then finished and curated by hand, image by image.
On hardware that meets our recommended spec, the game runs its language model on your own machine and plays offline. On lower-spec systems the same work runs over the network instead: on our servers or on a host you run yourself. You choose which.
The base game pairs a small local language model with Retrieval-Augmented Generation. The model does not write the story. It acts as a context-aware fill-in-the-blank engine over content that was already authored. During dialogue, the engine retrieves a human-written response from a pool of a hundred or more per character, and the local model adapts it to name your character, reflect your condition, and match how you phrased the question. In combat, pre-written action descriptions are updated to reflect your equipped weapon, your enemy, and current health.
The optional Sandbox World Generation Wizard is where players decide for themselves. You can build a world with your own writing and your own art, or bring in generative AI at whatever level you choose. Those tools can run entirely on your own hardware or on our servers.
Every generative path is bound by hardcoded prompt construction, local deterministic blocklists, trajectory tracking that resists multi-turn jailbreak attempts, and real-time safety classification.
Shared worlds carry a label describing how they were made: original writing and imagery, AI used to edit the author's own work, or AI used to complete the world. You see that label before you choose a world to sail.
IMMRAMA: Voyages of the Otherworlds
Gameplay Trailer
