How to write a more usable character card — structure, anti-patterns, revision order
“Usable,” for me, means fewer model detours and fewer rewrites for you. Everything below is operational.
Suggested structure (stable → detailed)
- One-line hook — what you’ll remember first about them.
- How they talk — two example lines beat a page of adjectives.
- Motivation and soft spot — one line each is enough for long arcs.
- Default scene — time, place, relationship—cuts random branching.
- Boundaries — what shouldn’t happen; write it plainly.
It’s fine if the first draft doesn’t fill everything—make (1) and (2) solid first.
Anti-patterns
- Everything in one blob — hard for humans and models to scan; painful to patch.
- Contradictory rules — “never X” in one block, “must X” in another.
- Traits without behavior — “easygoing” with no example of how that sounds.
Revision order (verbs)
- Read the whole card and mark the two spots that break immersion fastest in chat.
- Change only those two—don’t rewrite the world in the same pass.
- Save a new version, import, and test-chat.
- Repeat until drift is mostly plot, not persona.
Same idea as small code commits: one giant edit, and you won’t know what broke.
A private list, not a public directory
If you juggle many characters, keep a table: name | one-liner | path. You don’t owe the internet a master catalog—just find your own files.
See also
On macOS, revise, preview, and save Silly Tavern–style PNG character cards with Sillycard. Cheap iteration is what makes “better cards” realistic. Features: App Store and in-app copy.