Anatomy of a character an AI can actually play
Most character platforms are an upload button and a million half-written sheets. A character who stays themselves for two hundred scenes is not uploaded — they are built. Here is what the building takes.
Why most AI characters go limp
Open any character platform and you will find millions of uploaded sheets, and the same failure inside most of them: a pile of adjectives. Warm but guarded. Fiercely loyal. Dry sense of humour. Every word true, and none of them tell a model what to do when you say something careless at two in the morning.
A model performing "guarded" has to guess what guarded looks like this second — and it guesses the average of every guarded character ever written. That flatness gets blamed on the model. It usually started in the sheet.
This is why an upload button, by itself, produces a catalogue rather than a cast: writing a character who performs is craft, and craft is unevenly distributed.
A character is a set of decisions
What performs is not qualities but decision rules — a small number of them, each concrete enough to act on.
A want with a cost. Not "wants to be happy" but "wants to sell the workshop before winter, and cannot say so in front of her father". Now every scene in that workshop has a second layer. If getting the want costs nothing, the character is finished by scene two.
A line they will not cross. The most skipped element and the most useful one. "She will not talk about the year she was away." "He does not lie to children, even when it would help." Refusal is where a character becomes visible — anything the story pushes at those lines produces friction, and friction is where character shows.
A voice with mechanics. "Speaks bluntly" is a label. Performable is: short sentences under stress, answers a personal question with a question, one recurring physical habit. Specifics carry presence further than a paragraph of description.
A few facts that can each produce a scene. A person they owe. An object they keep. A place they avoid. Three or four hooks beat a chronological biography, most of which never becomes actionable.
A worked example
Before: Mira, 34, a talented restorer. Warm but guarded, with a dry wit. Loves old things and hates being pitied. Has a complicated relationship with her family.
After: Mira, 34, restores furniture in her father's workshop. Wants: to sell the workshop before winter. Cannot: say so while he is still in it. Will not: discuss the year she was away — deflects with work. Voice: short sentences, shorter when hurt; answers a personal question with a question. Habit: runs a thumbnail along a joint while thinking. Facts: owes money to a man called Teo; keeps her mother's chisel and does not use it; can tell the age of a piece by smell and never mentions it.
Same person. The second one can be performed — by any capable model, in any system. That much is craft, and craft is portable.
The part a sheet cannot carry
But even the good sheet has a ceiling, and it is worth naming, because it is the ceiling most AI characters live under.
A person is not only their decision rules. They are their history — the year that is not discussed actually contains something. They are their relationships — Teo is a person with his own stake in the debt, not a name in a field. They are their place in a world that knows them back. A sheet performed in isolation, however well written, is a monologue wearing a face.
That depth cannot be uploaded in a minute. It has to be worked out — the background that holds together, the ties between characters that pull in different directions, the details that will matter in scene forty seeded long before scene forty arrives.
Every character here is built, not uploaded. A page of decisions, a history that holds together, people they already belong to — and then the story finishes the person.
Founder, projectDigo
How projectDigo builds its cast
This is why projectDigo's characters are authored, not user-generated. Every character in the catalogue is built by hand, to the discipline above and past it: a worked-out background, real relationships to the other people of their story, limits that hold, a voice of their own — and an approved face they keep.
A curated cast is a deliberate trade. An upload button would fill the catalogue a thousand times faster; it would also fill it with adjective piles. We would rather ship fewer people who hold up for two hundred scenes than a million who go limp in ten.
And the craft ends the way good character writing always ends: by leaving room. The spine is built; what a character does under pressure becomes part of who they are, and that version is always better than one anyone could have planned. Stories are given the same treatment — their beats, their places, their stakes are worked out with the same care, but that deserves a piece of its own.
Questions we get
Why do adjectives fail in an AI character description?
Because an adjective names an impression without saying how to produce it. A model performing "guarded" improvises the average of every guarded character in its training data, and the result reads generic. A decision rule — she answers a personal question with a question — produces the same impression, consistently, because it says what to do.
What should a character description contain instead of adjectives?
A want with a cost attached, one or two lines the character will never cross, a voice described as observable mechanics rather than labels, and a handful of concrete facts that can each generate a scene. That spine outperforms a page of biography in any AI system.
Can I create my own character in projectDigo?
The cast is authored. Every character is built in-house — background, relationships, limits, voice and an approved face — because that depth of preparation is exactly what makes them hold up over a long story, and it is craft work rather than a form to fill in. What you get as a player is a world whose people were finished before you met them.
How does a character stay the same person over hundreds of scenes?
Two commitments working together: what happens to them is recorded and becomes part of who they are, so scene two hundred knows what scene three established — and their appearance is anchored to one approved image, so the face you met is the face you see. Change still happens, but it happens because the story caused it.
