A portable character card should describe your fictional companion without depending on one app. Keep an editable copy outside the chat so a new service, lost conversation or changed model does not require starting the character again.
Write the stable core
Give the character an explicitly adult age, a name, three personality traits and a preferred way of speaking. Describe what the character notices and what they avoid. A sentence such as “Mika is a 29-year-old kitsune who asks thoughtful questions and dislikes rushed decisions” gives a clearer starting point than a long list of flattering adjectives. Keep real addresses, workplace details and private information out of the card.
Separate identity from the current scene
Put permanent traits in one section and the current setting in another. A rainy bookstore belongs to the scene; a habit of finding humour in small mistakes belongs to the character. Keep a third section for boundaries such as staying within a fictional setting and not impersonating real people. This separation makes it possible to change the scene without changing the personality.
Run a small transfer check
Start a fresh conversation in the destination app, paste the card and ask for a short opening. Check the age framing, voice and scene rather than expecting identical words. If something changes, adjust one instruction and try again. Keep the version that expresses your intent most clearly, and label your local file with a revision date. Different services interpret prompts differently; portability preserves your instructions, not a guaranteed response.
Keep an exit copy
Save the card as plain text and keep any approved visual reference separately. Do not rely on a screenshot as your only editable copy. Before uploading the card to another service, re-read it for personal details and check that service’s current account and upload terms.
A complete example card
Here is an original example you can adapt: “Mika is a 29-year-old fictional kitsune who repairs old radios above a riverside bookshop. She has copper hair, dark fox ears and a blue wool coat. She is curious, patient and gently funny. She asks one question at a time and notices small details in her surroundings. She speaks in short, natural paragraphs and does not end every reply with a question. Her world is fictional. Do not borrow identifying details from the user’s real life or impersonate a real person. The opening scene is a rainy evening after the bookshop closes. Mika has found a radio that plays a different station whenever the rain stops.”
Why the example is organized this way
The first sentence establishes adulthood, identity and a setting that can support several stories. The appearance sentence is deliberately brief; a conversation prompt does not need a catalogue of visual adjectives. The next sentences describe observable behaviour rather than promising that the character will be perfect. The boundary sentence explains how to keep the fiction separate from real people. Finally, the opening scene creates something to respond to. You can remove that last sentence and use the same character at a train station or a café without changing the core card.
Adapt the card to a different interface
Some apps offer separate boxes for personality, greeting and scenario. If so, put each part in the box whose label matches its purpose. If the service gives you one prompt field, use short headings inside the text. Do not put the entire card into a public display-name field or an image caption simply because that is the first editable area you see. Inspect the visible profile after saving to check what other users may see. Keep any private instructions out of public fields, and consult the app’s current help if visibility is unclear.
A transfer test you can repeat
Ask the destination character three small questions: what they were doing before the scene began, how they would handle an interrupted plan, and what they want to investigate next. You are checking whether the answers fit the card, not whether the model repeats its text. Record one sentence about each answer. If the character invents a conflicting occupation, clarify the occupation once. If the tone is too elaborate, shorten the style instruction. Change only the part connected to the mismatch so you can tell whether the edit helped.
Troubleshoot without adding a wall of rules
A common mistake is to respond to every odd reply with another permanent instruction. Over time the card can contain several incompatible versions of the character. Before adding anything, read the existing text for duplication. Replace a vague phrase with a concrete behaviour and remove the outdated phrase. If the interface has a short field limit, preserve adulthood, core identity, two behavioural traits and the current scene first. Details such as favourite foods can live in your private notes until a scene needs them.
Keep a small version history
Name the file with the character and revision date, and add a one-line change note. Keep a stable version you enjoyed before experimenting. Store the text in a format you can read without the original app. An exported chat may help you remember a scene, but it should not become the only record of the character’s instructions. Before transferring again, remove unnecessary personal details and check the destination’s current terms. This method gives you control over your creative brief while leaving the service’s actual memory and privacy behaviour as separate questions.