AI and ethics

Every piece of content in Vivos is generated by AI. We think you should know which models we use and what we know about how they were trained.

How Vivos uses AI

The models below generate the words, images and music. They don't decide what happens. That's Vivos's engine. Each character has a personality profile and an emotional state that shifts with what happens in the story. The engine tracks what characters want, what they remember and how they feel, and decides what they do next. A set of directors watches every story as it unfolds: one introduces twists, one lays out each scene as a comic, and one turns your session into a book you can read later. The models give that system a voice and a look. Your story comes from the system.

Audio
Stable Audio, trained only on licensed music and Creative Commons recordings, with artists paid through a revenue share. This is the standard we want everywhere.
Images
FLUX models from Black Forest Labs. Their newest model, FLUX.3, is trained on licensed data with no web crawling, and it's available in Vivos.
Text
Gemma from Google, with Mistral Small 4 as a fallback. Like most language models today, both are trained largely on public web data.

We choose models on three things: what they can do, how well they run, and how their training data was sourced. This field moves fast, so the models we use will change. When a model with better sourcing meets our bar on the other two, we switch, and we update this page when we do.

When AI gets it wrong

We work to keep mistakes rare, but AI still gets things wrong: a character says something that doesn't fit, or an image comes out with six fingers or a melted face. Models are improving fast, but we won't pretend the problem is solved.

So here's our commitment: whenever you create something new, whether an actor, a location or an object, you can regenerate its visuals at least once for free. If it still isn't right, contact us on Discord or at contact@duplicants.gg, and we'll fix it ourselves.

Longer term, we're building a way to flag bad content directly in Vivos, so fixes happen faster and help improve the system for everyone.