← All features

Character Training

The same face, every single time

Generate enough art of a character and you can train a model on it. From then on their face is theirs — not a fresh interpretation of a prompt every time you ask for a picture.

How it works

  1. 1

    Generate art of your character the ordinary way until you have a set you like.

  2. 2

    Choose which of those images the training run should learn from — at least ten.

  3. 3

    Training runs in the background; you keep playing while it works.

  4. 4

    Once it finishes, new images of that character are generated through the trained model automatically.

  5. 5

    Retrain whenever their look changes, picking a different set of reference images.

You choose the training images — not whatever happened to be most recentMinimum of ten reference images per runTraining runs in the background, queuedApplies automatically to new generations of that characterRetrainable as a character's appearance evolvesWorks alongside the equipment and visual-state systems, not instead of them

Why the image picker matters

A trained model is only as good as what it learned from. Feed it twenty near-identical shots of the same campfire and it learns the campfire, not the character — so the selection is yours to make deliberately, rather than being taken automatically from your most recent generations.

Ready to try it yourself?

Start Adventure →