If you've spent time in SillyTavern, you probably have a folder of .png character cards you've collected or written yourself. Most places you can take them are text-only — the card stays a card. ximages does something different: import a card, and the character gets a real photoreal face (same one in every image, via My Character) plus an actual chat you can hold with it.
The short version
- Open My Character → Create a character → Import Tavern card
- Upload the
.png or .json file
- Review the name, description, and opening line it pulled out — edit anything that reads better for an image prompt than it did for pure text roleplay
- Generate a reference face, pick the one you like, save
- Hit chat — if the card had an opening line, the character says it first, same as it would in SillyTavern
Free to import and create. Generating scenes or chatting past the free tier uses credits, same as everything else on the site.
What actually gets read from the card
SillyTavern cards are a JSON blob — either embedded in a PNG's metadata (V2 spec uses a chunk called chara, V3 uses ccv3) or shared as a plain .json file with the same fields. ximages pulls:
| Card field | Becomes |
|---|
name | Character name |
description + personality + scenario | Combined into the appearance/persona text (used for both the reference face prompt and the chat system prompt) |
first_mes | The character's opening line — shown (and actually saved, not just displayed) the first time you open chat |
mes_example and anything V3-specific (assets, alternate greetings, lorebook) aren't used yet — the core fields above are what drive both image generation and chat today.
One honest caveat
Cards written for pure text roleplay often describe personality and backstory in detail but skip the kind of concrete visual detail a good image prompt needs — hair color, eye color, build. If your imported character's reference face comes out generic, that's why: add a sentence or two of physical description to the appearance field before generating. The import gives you a draft to edit, not a finished product — nothing is created until you review and hit save.
Also worth knowing: card content goes through the same moderation every other prompt on ximages does before it's used. A card that doesn't pass won't import.
Why bother, if the card already works in SillyTavern?
Because SillyTavern chat is text-only. There's no face. ximages adds the part most roleplay tools don't do: a locked, consistent, photoreal face for the character via IP-Adapter reference-image conditioning — generate her in any scene, outfit, or pose, always the same person — on top of a real chat, not a static sheet.
FAQ
Does this support V2 and V3 cards?
Yes. V2 (chara_card_v2, the format every frontend reads) and V3 (ccv3) both parse — V3's extra fields beyond the shared core aren't used yet, but the character itself imports fine.
PNG or JSON — does it matter which?
No, both work. PNG is the more common way cards get shared (the JSON is hidden inside the image); plain .json exports work identically.
Does importing modify or upload my original card anywhere public?
No — the file is parsed server-side to extract text fields, nothing about the original file is published or shared. What becomes public (if anything) is only the character you create from it, and only if you explicitly set it to public.
Can I still use the same card in SillyTavern after importing it here?
Yes, nothing about the source file changes. Import just reads it.
Do I need an account?
Yes — importing and creating characters requires being logged in (an anonymous trial account works too).
Import your first card →
Related: How to make a consistent AI girlfriend (same face every image) · Which AI generators actually keep the same face? · Best AI girlfriend generators in 2026