About
Contact and background
Why this exists
Character cards move between tools constantly, and the tools disagree. One writes V2 flat, the next
expects the wrapped form, a third only reads the chara chunk and ignores ccv3.
When a card fails to load somewhere, there is usually no way to see what is actually inside the file.
Larkbench is the inspector we wanted: open the file, see every field and every worldbook entry as they really are, fix what is wrong, write it back out in the shape the destination expects. And because a card is just metadata inside an image, none of that requires a server — so it does not use one.
Who
Larkbench is built and maintained by a small independent software team. We work on tooling for conversational-AI workflows; this is the piece we could make useful on its own, so we published it.
Contact
Bug reports, format edge cases, cards that fail to parse, or anything about the privacy policy or terms:
If a card will not parse, telling us the detected version and the error text is usually enough — please do not send the card itself unless we ask. We would rather not receive files we have gone to some trouble to avoid receiving.
Roadmap
Reasonably likely, in rough order: unpacking V3 chara-ext-asset_ chunks,
standalone worldbook (WorldInfo) files, preset inspection, and batch conversion of several cards at once.
All of it stays local — anything that would require uploading your files will not be built.