1. Upload modern Card or JSON
Technical Formatting: Indents and Structure
KoboldAI users often manually edit their JSON files to tune model performance. Our converter aids this by providing a clean, indentation-aware output:
1. Indentation Preservation
One of the most common issues with automated converters is the loss of whitespace. Many local models use indentation to distinguish between different character traits. Our TinyLib parser ensures that multi-line personas remain formatted correctly for Kobold's context window.
2. Memory vs. World Info
Kobold divides its understanding between "Memory" (permanent personality) and "World Info" (context-triggered lore). Our converter places the persona and scenario data into the fields that Kobold's Lite UI reads as the character's core memory block.
3. Metadata Stripping
SillyTavern PNGs often contain 10-20KB of "UI-Only" metadata (like custom character colors or folder paths). Kobold does not need this. We strip this extra weight to ensure your context window is used only for character definition and conversation history.
How to use in KoboldAI Lite
- Convert your card and download the
.jsonfile. - Open KoboldAI Lite in your browser.
- Go to the "Settings" or "Characters" tab.
- Click "Import JSON" and select the file you just saved.
- Your character's name, description, and greeting will automatically populate the relevant fields.
FAQ: KoboldAI & KoboldCPP
A: Yes. KoboldCPP is the backend, and if you use it with the built-in Lite UI, this JSON format is exactly what you need to load your characters.
A: While Kobold Lite supports some PNG cards, raw JSON is 100% reliable across all versions of the software and is easier to edit if you need to manually tweak a character's traits.
